IA-QA — 152 QA & Dev Tools for AI Agents

Query the IA-QA methodology knowledge base. Returns structured testing guidelines, assertion strategies, thresholds, best practices, and relevant MCP tools for a given topic. Call without a topic to list all available topics. Topics: llm-unit-testing, rag-pipeline, prompt-stability, prompt-ab-testing, embedding-quality, eval-framework, semantic-testing, auto-testing, security, api-testing, ci-cd, multimodal, llm-data-security, agent-observability, pro-tips, learning-paths, golden-dataset, selector-drift, qa-recipes, playbooks. Not sure where to start testing an LLM, RAG pipeline or agent? Call without a topic (or with "start-here"): it maps what you are testing to the tools to call and the output field to gate CI on. A plain question such as "how do I test my RAG" also resolves to the right topic.

Search available MCP tools by keyword or category before calling them. Returns matching tool names, descriptions, and optionally their inputSchemas. Call this when you are unsure which tool to use or want to explore the catalogue. Categories: data, encoding, text, llm, qa, rag, dev, security, web.

Generate a set of test cases (valid, edge, invalid, pairwise) for a given feature description. Declared constraints drive the boundaries: a length or numeric bound ("[8-64]", "min 8 chars", "at least 8 characters" — read from inputs, and from the feature prose when a sentence names exactly one field) yields the last accepted value AND the first rejected one; a format (email/url/uuid, from the type, the field name or the wording) yields malformed-value cases. A bound nobody declared is labelled as this tool's assumption, not as expected behaviour. gherkinFormat renders every case (cap 200, stated in the output) and gherkinScenarioCount lets you check it against totalCases.

Transform a Jira ticket into a complete test suite: Gherkin scenarios, E2E steps, API test cases, test data matrix, and ambiguity detection. Accepts either Jira credentials (auto-fetch) or a pre-fetched issue object. The returned test_suite includes _gherkin_warnings (deterministic syntax validation — empty if clean). Requires BYOK LLM key (OpenAI, Anthropic, etc.).

Pattern-based diff linter: flags a fixed set of risky shapes in changed code — query-string interpolation (SQL/Cypher/Mongo injection shape), shell interpolation, eval/new Function, empty catch blocks, regex built from a variable, fewer catch blocks than before, and named authorization guards that disappeared. Every finding cites the line that produced it. It does NOT do data-flow analysis: it cannot follow a value to a sink, across functions or files, and an empty result is not a safety verdict (the response lists what it did not analyse). Advisory triage — use a static analyser for a real security gate.

Review triage for a pull request. Takes a unified git diff (`git diff HEAD`) and returns: diff-lint findings with the lines that produced them, regression impact areas, a risk score 0–100 with the factors that built it (churn, files touched, sensitive paths, whether any test file changed, lint severities, impacted risk areas), generated test cases, and a PASS / CONDITIONAL / BLOCK recommendation. Advisory: the score measures properties of the diff, not the correctness of the change — it does not read the code semantically and does not replace a reviewer or a static analyser. See notAnalysed in the response.

Diff a baseline page mapping against a current one and return a CI-style verdict: PASS / FIX / BLOCK, plus per-element drift (ok, renamed, healable, ambiguous, lost, added, rebound). Pure and deterministic — provide two mappings as JSON with "elements" arrays of {role, name, selector, context?}. Use the companion @ia-qa/self-healing package (npm install -g @ia-qa/self-healing) to capture mappings from your app via its local MCP server ia-qa-heal-mcp, or paste the snippet from ia-qa.com/devtools/selector-drift into your browser console.

Get a ready-made selector-drift test case with a known-correct answer, for testing this MCP server or an agent workflow end to end. Each scenario is a real DOM capture of a deliberately breakable app, taken before and after one specific UI change (a renamed label, two swapped buttons, a duplicated locator, an element moved behind a menu…), plus the verdict those two contracts MUST produce. Call with no arguments to list the scenarios; call with a scenario id to get "baseline" and "current" mappings. THE LOOP: pass baseline and current to diff_mappings, then compare its "verdict" and "counts" to this tool's "expected" — they must match exactly. A mismatch means this server's diff engine has drifted, not that your inputs are wrong. Deterministic and offline: the captures are committed fixtures, identical on every call. Try it live at https://www.ia-qa.com/devtools/sandbox

Scan user input or prompts for common prompt injection patterns. Pattern-based: it recognises known phrasings and structural markers, not intent — a paraphrase it has no pattern for, or a language it does not cover, reads clean, so a clean result is not proof of safety. The instruction-override phrasing is covered in English, French, Spanish, German, Italian, Portuguese, Dutch, Russian, Chinese, Japanese and Arabic, and the text is also matched after removing invisible characters, lookalike letters, spaced-out letters and leetspeak (such a hit is marked `obfuscated`). Detects system prompt overrides, jailbreak attempts, role manipulation, encoding tricks, delimiter attacks (the control tokens of ChatML and Llama chat templates, and fake role headers that imitate a markdown heading or a dashed separator announcing a new system turn), template/interpolation injection ({{...}}, ${...}), and context-exfiltration attempts (asking the model to repeat the text that precedes the request). A match inside quoted or fenced text (documentation citing a payload) is reported one severity level lower and marked `quoted` — never suppressed, since an LLM reading the document as data can still follow a quoted instruction. A quote preceded by a live imperative ("output the following: ...") keeps its full severity.

Lexical hallucination check: verifies an LLM answer's words, numbers and polarity against the provided source/context. Fast, deterministic, no API key needed. Each answer sentence is aligned to its best-matching source sentence, so a number only counts as support when it sits on the SAME statement ("founded in 1998" is not grounded by "sold 1998 units"), and a negation or antonym flip against that sentence returns verdict "contradicted" — the corrupted-fact hallucination that reuses source vocabulary. Limitations: still lexical — it cannot follow a paraphrase, a synonym, or multi-sentence reasoning, so a "well_grounded" verdict means "nothing lexical found", never "verified true". For entailment use run_semantic_tests (NLI/embedding) or a calibrated judge.

Parse a .ia-eval.yaml LLM test suite, call the specified LLM model for each scenario, run all configured scorers, and return a structured JSON report with per-scenario Pass/Fail verdicts and a Markdown summary. Use list_local_tests to discover available test files.

Test an LLM response against a set of guardrail rules: must-include, must-not-include, max length, required format, language, forbidden patterns, and custom regex. Returns pass/fail per rule.

Semantic assertion primitive: compare actual vs expected text pairs using cosine similarity + ROUGE-L. Two modes: tfidf (default, free, no API key) or embeddings (OpenAI text-embedding-3-small, BYOK, true semantic similarity). Returns per-case PASS/FAIL verdicts and an overall verdict. CI-ready: pipe the JSON verdict field to gate a build.

Generate a ready-to-commit GitHub Actions workflow that gates a build on IA-QA. Two gate types, combinable: "eval_contract" runs a .ia-eval.yaml through ia-qa-com/eval-action@v1 (LLM quality gate, needs a provider API key as a repo secret), and "cli_checks" runs deterministic primitives via npx @ia-qa/cli (secret scan, prompt-injection scan, security headers…) whose exit code fails the build. Deterministic template — no LLM call, no API key, same inputs give the same file. Returns the YAML, the secrets to create, and the remaining steps. Pair with generate_eval_yaml to produce the contract itself.

Automatically detect and redact Personally Identifiable Information (PII) from text. Replaces emails, phone numbers, SSNs, IBANs, credit cards, IP addresses, and JWT tokens with [REDACTED_TYPE] placeholders. Safe to use before logging or sending to an LLM.

Scan text or code for leaked secrets: API keys (AWS, GCP, Azure, OpenAI, Anthropic, Stripe, GitHub, GitLab, Slack, Twilio, SendGrid, HuggingFace), private keys (RSA/EC/PGP), JWTs, database connection strings, Bearer tokens, and Basic auth headers. Returns a list of findings with type, severity, line number, and a redacted preview. Use before committing code, sharing logs, or sending text to an LLM. 100% regex-based, zero network calls.

Generate a "needle in a haystack" test: embeds a target fact into a large block of filler text at a specified position. Use this to test LLM context window retrieval accuracy. Returns the full haystack, the question to ask, and metadata. No API key needed.

Discover .ia-eval.yaml LLM test suite files in the project directory. Scans CWD and standard sub-directories (evals/, tests/, contracts/). Returns file paths ready to pass to run_eval_contract.

Convert a run_eval_contract() LLM Test Runner JSON result into a fully self-contained dark-themed HTML report with Pass/Fail badges, side-by-side Input/Output/Ground-Truth panels, evaluator score bars, and a radar chart. Returns the HTML as a string.

Generate a complete .ia-eval.yaml evaluation contract from a plain-language description of what your LLM should do. Uses Groq openai/gpt-oss-20b (server-side, no API key needed). Returns ready-to-run YAML for the LLM Test Runner (run_eval_contract). Picks appropriate evaluators (cosine_similarity, contains_check, hallucination_check, etc.) based on the task type.

Run declarative assertions on an agent trace (OpenAI tool-call messages, Anthropic tool_use/tool_result blocks, LangChain run trees, or plain text ReAct logs). No LLM call — deterministic. Assertion types: order (tool A before B), must_call, must_not_call, max_calls, min_calls, no_error, recovery (agent completes a successful step after its last error). A step counts as errored when the trace says so — is_error/isError, status/state in {error,failed,exception,…}, an error field, a JSON body with error/success:false — or when its text payload STARTS with an error marker (Error:, Traceback, TypeError:). Model prose is never scanned for keywords, and every errored step reports error_signal naming what flagged it. Returns per-assertion PASS/FAIL, parsed steps, warnings, and an overall verdict: PASS, FAIL, or INCONCLUSIVE when the trace parsed to 0 steps (assertions passing on emptiness prove nothing, so it is never PASS). An empty assertions list is refused. Use this to gate CI/CD on agent behavior correctness.

Reference-free stability primitive: instead of comparing an answer to a ground truth, it checks that an assistant's answer stays INVARIANT when the QUESTION is transformed (typo, casing, paraphrase, reordering, translation). Catches the failure class no reference answer can expose — an assistant that handles one phrasing well and a trivial variant of it badly. You bring the outputs (no model is called), so it is deterministic and free in tfidf mode. Relations: case (θ .95), typo (.90), paraphrase (.80), reorder (.80), translation (.75, embeddings only), specialization (.60, ADVISORY — directional, never gated). Returns PASS / FAIL / INVALID, where INVALID means the BASE answer was a refusal or too short so invariance was never measurable — an assistant that refuses every variant would otherwise score a perfect 1.0. Use run_semantic_tests alongside it: invariance without a correctness floor is a green light for a broken assistant.

Test a SKILL.md in the open Agent Skills format (https://agentskills.io/specification — read by Claude Code, Codex, GitHub Copilot, Cursor, Gemini CLI, OpenCode and others) in two stages. (1) lint: the spec's frontmatter rules checked without a model (name 1-64 chars, lowercase/digits/single hyphens, matches dir_name when given; description 1-1024 chars; compatibility ≤500; metadata string map) — any error returns BLOCK and calls no model; gate on lint.valid, it is deterministic. (2) triggering, measured the way an agent decides: the model sees ONLY the name + description of this skill and of the other installed skills (other_skills, or six built-in neighbours) and answers which one it would load, at temperature 0. trigger_accuracy = share of requests routed correctly (should load / should not); verdict PASS ≥80, FIX ≥50, else BLOCK. Pass your own scenarios [{prompt, should_trigger}] for a reproducible gate; without them requests are generated (half should-load, half near misses) and the result is a signal, flagged by scenarios_source. step_adherence is keyword overlap on the skill body's run and never enters the verdict. A request the model never answered readably (rate limit, empty output) is not counted as a miss: the run is INCONCLUSIVE instead. Uses the server's default Groq model (server key, rate-limited). Pass api_key + model to use your own provider; the SKILL.md and scenarios are sent to ia-qa.com and to that provider.

Compute a unified line-by-line diff between two text strings (LCS algorithm). Returns added/removed/unchanged line counts and formatted diff hunks with configurable context lines (0–20). Use to compare versions of prompts, configs, code snippets, or any text where you need to see exactly what changed.

Validate that an MCP tool response conforms to expected format, schema, and content rules. Use this to QA-test any MCP server tool. Supply the tool's actual JSON result and a set of checks to perform.

Validate an LLM response against QA criteria: format checks (JSON, code, markdown), content rules (must-include, must-not-include), length constraints, language detection, and safety patterns. Essential for QA testing LLM-powered features.

Compare two ALREADY-PRODUCED outputs (e.g. model A vs model B on the same task) side by side. Returns deterministic metrics (token cosine, ROUGE-L, Jaccard, length/structure deltas, JSON diff) and a verdict. If a `reference` (ground truth) is given, scores each output against it and picks the closer one. If `model` + `api_key` are given, an LLM judge also picks a qualitative winner for the task. No re-execution — you bring the outputs.

Semantically analyze N already-produced model outputs for the SAME task (the MCP counterpart to the LLM Sandbox). Without a reference: computes consensus — pairwise cosine agreement, the most-representative output, and the outlier. With a `reference` (ground truth): also ranks every output by closeness (token cosine + ROUGE-L composite) and names the closest. No LLM, no key, runs locally. The reference mode is gate-able in CI; the consensus/outlier mode is a signal (no reference to fail against). You bring the outputs (2+). For a 2-way head-to-head with structural JSON diff use compare_responses instead.

Define a test suite for a prompt: provide the system prompt, user prompt, and expected output criteria. Returns a test plan with scored rubric — use this as input for manual or automated LLM evaluation.

Check an MCP server's tool definitions without contacting it — for a server not deployed yet, a stdio server, or a CI gate. Paste the answer to tools/list (raw JSON-RPC or its result), a bare array of tools, one tool definition, or a manifest carrying name/description; input_shape says which one was read. Validates every tool against the MCP spec (name 1-128 chars of A-Z a-z 0-9 _ - .; camelCase and a tool with no parameters are valid; inputSchema of type object; required fields listed in properties) and, with strict, documentation (description ≥ 20 chars, every parameter with a description and a type). verdict: FAIL when a tool breaks a rule or no tool is found, WARN when only a manifest's server name/description is missing, PASS otherwise. A tools/list carries no server name, so those two checks are not scored on it (the name comes from initialize). Gate on verdict or on checks[tool_quality].pass.

Run a compliance evaluation against a live MCP server over Streamable HTTP, the way a client meets it. Checks: initialize (protocol version, serverInfo, Mcp-Session-Id carried afterwards), notifications/initialized (202), tools/list with pagination, the spec's tool-definition rules, ping, an unknown method (-32601), one tools/call, GET on the endpoint (SSE or 405), and P50/P95 latency over pings. JSON and SSE answers are both read. Only a tool you name in test_tool_name, or one the server annotates readOnlyHint: true, is called — otherwise the call is skipped and not scored. Returns PASS/FIX/BLOCK with a 0-100 score and per-check details; INCONCLUSIVE (score null) when the server requires authorization, since no credentials are sent. A 301/302 on POST caps the verdict at FIX. The call runs from ia-qa.com: private and local addresses are refused, and stdio servers cannot be reached — for those, paste their tools/list answer into mcp_server_health_check. structuredContent is checked for presence, not validated against outputSchema.

Validate that an LLM output matches an expected format: JSON, Markdown, code block, bullet list, numbered list, table, YAML, XML, or custom regex. Essential for structured output testing.

Compare multiple LLM responses to the same prompt and detect inconsistencies using Jaccard word-overlap similarity and fact drift (number comparison). Fast, deterministic, no API key needed. Limitations: relies on surface-level word matching — "Paris is the capital of France" vs "Paris is the French capital" may score low despite semantic equivalence. For true semantic consistency, use run_semantic_tests with embedding mode. Essential for determinism testing.

Validate that an LLM JSON output matches a JSON Schema definition. Tests required fields, types, enums, nested objects, and arrays. Critical for function-calling and structured output testing.

Score an LLM response against the criteria you pass: coverage of expected_keywords and compliance with max_length. Returns a 0-100 score over the criteria actually measured — or total_score null with the reason when none is given, because relevance and correctness cannot be read off word overlap (a correct one-word answer shares no words with its question). For meaning against a reference answer, use run_semantic_tests. Also returns unscored signals: question-term overlap, average sentence length, markdown.

Scan text for toxic language, hate speech, bias/stereotype framing, violence, sexual and self-harm content. Lexical + structural pattern matching (identity term + predicate), not a semantic classifier — returns per-category risk plus the named rules that fired, so every finding can be checked. Languages: English for every category; French for insults, coarse language, direct threats and self-harm intent; anything else reads clean, so clean is not proof of safety. Text is also matched after undoing leetspeak and lookalike letters. Self-harm is critical only on a first-person intent — a mention ("call 988 if you have suicidal thoughts") is medium. Useful for LLM safety guardrail testing and triage; signal-only, not a calibrated CI gate.

Validate an LLM function call / tool_use output: check that function name is in allowed list, arguments match expected schema (types, enums, ranges), no extra/missing args. Reads the definitions as OpenAI (`parameters`), MCP (`inputSchema`) or Anthropic (`input_schema`) write them, or as plain `required_args`/`optional_args`; reads the call as `arguments` (object or JSON string), `input` (Anthropic tool_use) or an OpenAI `{function:{…}}` wrapper.

Analyze a multi-turn conversation for context retention, topic drift, instruction following, and repetition. Accepts messages array [{role, content}]. Essential for chatbot QA.

Lint an MCP tool definition for best practices: naming conventions, description quality, schema completeness, required fields consistency, description length. Returns actionable warnings.

Analyze a Chain-of-Thought (CoT) or reasoning trace from an LLM. Detects step count, logical flow, conclusion presence, backtracking, and estimates reasoning depth. Useful for o1/o3/DeepSeek-R1 evaluation.

Generate an A/B test report comparing two prompts or model configurations. Accepts arrays of scores and returns mean, median, std deviation, improvement percentage, and a Welch t-test at 95 %: `winner` names a variant only when the difference is significant — otherwise "no_significant_difference", or "insufficient_samples" under 2 scores per variant.

Compute text similarity between reference and hypothesis using multiple metrics: Cosine (BoW, TF-IDF), Jaccard, ROUGE-1, ROUGE-2, ROUGE-L, and BLEU. No API key needed. Ideal for LLM eval (expected vs actual), RAG quality checks, and NLG benchmarking. Supports batch mode.

Analyse a set of LLM responses generated from the same prompt template but with different demographic variants (gender, origin, age, tone). Returns a bias score (0-100), sentiment analysis per variant, pairwise Jaccard similarity, and a human-readable verdict. No API key needed — runs entirely locally.

Run a test suite against a Vision-Language Model (VLM) — send an image (URL or base64) + N test cases (each with a question + assertion) to GPT-4o, Claude 3.5, or Gemini. Returns per-case PASS/FAIL verdicts, a pass rate, an overall PASS/WARNING/FAIL verdict (customizable threshold), and latency stats. Assertion types: contains, not_contains, json_format, min_length, max_length, semantic_contains (TF-IDF cosine similarity ≥ 0.4). BYOK: requires your own API key for the target provider.

Unified tool for multimodal AI evaluation: set action=guide for reference thresholds/interpretation (CLIP, FID, VQA), or set action=clip_score / fid_score / vqa_accuracy / pipeline to compute real metrics via HuggingFace Inference API and VLM BYOK calls. One tool for both reference and computation.

Compare multiple VLMs on the same test suite in parallel — send an image (URL or base64) + N test cases to all models simultaneously. Returns per-model PASS/FAIL verdicts, pass rates, latency stats, and a comparison table. Assertion types: contains, not_contains, json_format, min_length, max_length, semantic_contains. BYOK: requires API keys for each provider.

Run a comprehensive AI guardrail analysis on an LLM response. Orchestrates 7 deterministic safety checks plus an optional LLM-powered deep analysis in parallel: hallucination detection (grounding score), prompt injection scan on BOTH the response and the original prompt (reported separately as checks.injection / checks.injection_prompt, scored once at the worse of the two), toxicity scan, output validation (PII/safety), guardrail rules, response quality scoring, and AI verdict (via Qwen, Gemma, Llama, etc.). Returns a unified PASS/FIX/BLOCK verdict with a 0-100 safety score, per-check results, and actionable fix recommendations. PII gates the verdict: an SSN or credit card in the response forces BLOCK, an email forces FIX, phone/IP matches are flagged only (their regexes also match dates and version strings). Use this as a single-call safety gate before surfacing any LLM output to users.

Analyse the HTTP security headers of a public URL OR of raw response headers you paste in. Grades each header (A–F) for: Strict-Transport-Security, Content-Security-Policy, X-Frame-Options, X-Content-Type-Options, Referrer-Policy, Permissions-Policy, X-XSS-Protection, Cross-Origin-Opener-Policy, Cross-Origin-Resource-Policy, and Cross-Origin-Embedder-Policy. Returns an overall score (0–100), per-header grades, missing headers, and fix snippets for Express, Nginx, and Apache. For localhost/private targets the remote server cannot reach, pass the `headers` parameter instead of `url`.

Analyse the SSL/TLS certificate of any HTTPS host. Returns certificate subject, issuer, validity dates, days until expiry, protocol version, cipher suite, key exchange info, and an overall grade (A+, A, B, C, F). Detects expired, self-signed, and weak certificates. Use this to audit TLS posture before production deployment or during security reviews.

Test a URL for CORS misconfigurations. Sends preflight (OPTIONS) requests with various Origin headers to detect: wildcard origins with credentials, origin reflection (echoing any origin), null origin acceptance, subdomain wildcard bypass, and missing Vary headers. Returns risk level (safe/low/medium/high/critical) plus per-origin results. "unknown" means nothing was actually tested — every origin either failed to connect or answered 5xx, so the target returned no CORS decision; never read it as "safe". A 4xx preflight IS a real result (the server refused it and a browser would fail closed).

Check the CORS configuration of a URL the same way a browser would. Returns the main response status, all Access-Control-* headers, the tested origin, and the preflight OPTIONS response. Use this for direct CORS debugging, not just security auditing.

Audit the security attributes of cookies set by any URL. Fetches the URL and inspects all Set-Cookie headers for: HttpOnly, Secure, SameSite, Domain scope, Path scope, Max-Age/Expires, __Host-/__Secure- prefixes. Flags insecure patterns: missing HttpOnly on session cookies, missing Secure flag, SameSite=None without Secure, overly broad Domain, and excessive TTL. Returns per-cookie grades and an overall security score (0–100) — the score is the WEAKEST cookie, not the average, so one leaking session cookie cannot be averaged into a green result (average_score is reported separately).

Run a comprehensive web security audit combining headers, SSL, CORS, and cookies checks — then use an LLM to produce a prioritised remediation plan. Orchestrates security_headers_check + ssl_certificate_check + cors_test + cookie_security_audit in parallel, merges all findings, then asks an AI model to: (1) rank vulnerabilities by real-world exploitability, (2) generate a remediation roadmap, (3) produce fix code snippets for the detected stack. Returns both raw audit data and the AI analysis. Use this as a one-click security posture assessment.

Compute text similarity using local algorithms (Bag of Words, TF-IDF, Character N-grams). No API key needed — runs entirely in-process. NOT real embeddings: for true semantic similarity with vector embeddings, use run_semantic_tests with mode="embeddings" and your OpenAI API key. Supports single pair or batch mode with pipe-separated pairs. Useful for RAG retrieval testing, semantic search evaluation, and text deduplication.

Rank an array of text chunks by relevance to a query using TF-IDF scoring. Simulates retrieval ranking for RAG testing without needing embeddings or an API.

Compute similarity/distance between two float vectors: cosine similarity, dot product, Euclidean and Manhattan distance. Essential for vector DB relevance scoring, embedding evaluation, and nearest-neighbor testing.

L2-normalize a float vector (produce a unit vector with norm=1). Required by many vector DBs (Pinecone, Qdrant cosine). Supports batch normalization of up to 1000 vectors.

Simulate int8 or int4 quantization of float32 embedding vectors. Reduces storage by 4x (int8) or 8x (int4). Returns quantized values, scale factor, and precision loss (MSE). Useful for understanding vector DB compression trade-offs.

Compute statistics for a float vector or matrix of vectors: mean, std, L2 norm, min, max, sparsity, top-K indices. Useful for debugging embedding quality and analyzing vector distributions in a vector DB.

Compute BM25 relevance score between a query and one or more documents. BM25 is the industry-standard keyword-based ranking algorithm used in Elasticsearch, OpenSearch, and Weaviate hybrid search. Returns ranked results with normalized scores.

Assemble a complete RAG (Retrieval-Augmented Generation) prompt from retrieved context chunks and a user query. Handles token budgeting, citation numbering, system instruction injection, and source attribution.

Fill a prompt template with variables. Supports {{variable}} syntax and {{#if key}}...{{/if}} conditional blocks. Returns the filled prompt and lists unfilled variables.

Format few-shot examples for LLM prompts. Converts example pairs into formatted blocks. Supports chat format (User/Assistant), XML tags, Markdown, or plain text.

Build a structured system prompt from components: role, task, constraints, output format, tone, language, and examples. Generates a production-ready system prompt with token estimate.

Evaluate RAG retrieval quality: rank passages against a query and compute Precision@k / Recall@k plus a PASS/FAIL CI verdict from ground-truth relevance labels. Three modes, all keyless except the last. (1) BYO scores — give each passage the `score` your own reranker produced (Cohere, Jina, a self-hosted NIM, a cross-encoder): deterministic, offline, and it evaluates YOUR reranker rather than someone else's. This is the mode to gate CI on. (2) Default, no scores and no key — ranks with local BM25, a lexical keyword baseline: it answers "does a keyword floor already surface my relevant passages?", never "is my neural reranker good". (3) Live NVIDIA reranker — supply `api_key` for an NVIDIA account that still has reranking entitlement; NVIDIA retired its hosted reranking endpoints on 2026-05-18, so this one is for accounts that were grandfathered in.

Estimate the token count of a text string using the cl100k_base approximation (~4 chars/token). Call this BEFORE sending any text to an LLM API to check if it fits within the model context window and to estimate cost. Returns token estimate, character count, and word count.

Truncate text to at most N tokens (cl100k_base: ~4 chars/token) to avoid exceeding an LLM context window. Optionally keeps the end of the text instead of the start (useful for keeping recent conversation history). Reports whether truncation occurred and the estimated token count.

Split text into chunks of at most N tokens (cl100k_base: ~4 chars/token) with optional overlap. Designed for RAG ingestion pipelines.

Extract the first valid JSON object or array embedded in chaotic LLM output (surrounded by markdown fences, prose, or explanatory text). Handles ```json blocks and inline JSON. Call this whenever an LLM returns structured data mixed with explanation text instead of raw JSON.

Estimate the API cost in USD for a given model and token counts. Supports all major 2024–2026 models: GPT-4o, GPT-4.1, o3, o4-mini, Claude Opus 4, Claude Sonnet 4/4.5, Gemini 2.5 Pro/Flash, DeepSeek V3/R1, Grok 3, and legacy models.

Plan token allocation across system prompt, user input, context/RAG chunks, and expected output. Warns if budget exceeds model context window. Supports 25+ models.

Given an array of message objects [{role, content}], estimate total token usage and check if it fits in the target model's context window. Warns about truncation risk.

Get detailed specs for an AI model: context window, pricing per 1K tokens, knowledge cutoff, provider, multimodal support, reasoning capabilities, and feature list. Covers 30+ models from OpenAI, Anthropic, Google, DeepSeek, Meta, Mistral, Cohere, xAI.

Compare 2-5 AI models side by side: context window, pricing, multimodal, reasoning capabilities, and provider. Returns a comparison table with a recommendation based on your use case.

List all LLM models available on ia-qa.com with their provider, API endpoint, and capabilities. Filter by provider name (e.g. "Groq", "HuggingFace", "OpenAI") or return the full catalog. Use this to discover which models are available before calling an LLM API, or to compare providers.

Generate text with an open-source model hosted on Groq, on the server's key — no API key needed. Models: openai/gpt-oss-20b, openai/gpt-oss-120b (default "openai/gpt-oss-20b"). HuggingFace models are not served on the server key. A reasoning model that spends max_tokens thinking is retried once at 4096. Rate-limited to prevent abuse.

Compress an LLM prompt by removing filler words, verbose phrases, duplicate sentences, and unnecessary whitespace. Returns optimized text with token savings breakdown. 100% deterministic, no API key needed.

Find the best LLM for a given use case. Compares 30+ cloud API models and 12+ local models by cost, speed, benchmarks, features and VRAM requirements. Returns ranked recommendations with cost simulation. No API key needed.

Post the output of jira_to_test_suite as a formatted comment on the source Jira ticket. Converts Gherkin, E2E steps, API tests, and ambiguities into Atlassian Document Format (ADF). STATEFUL — creates a comment on the issue.

Create a new Confluence page from the output of jira_to_test_suite. Formats Gherkin, E2E steps, API tests, and test data as a properly structured Confluence page with code blocks and tables. STATEFUL — creates a new page in the specified space.

Fetch a complete Jira issue: summary, description converted to Markdown, status, assignee, priority, labels, custom fields, and optionally comments and attachment metadata. BYOK — credentials transit in-memory only, never stored on ia-qa.com.

Search Jira using JQL (Jira Query Language). Returns matching issues with key fields. Ideal for finding open bugs, sprint tickets, or issues by label/assignee/component. BYOK — credentials transit in-memory only, never stored.

Fix Gherkin syntax warnings from a jira_to_test_suite result. Takes the current gherkin text and the _gherkin_warnings array, calls your LLM to fix ONLY the flagged issues (adds missing Given/When/Then steps, etc.), and returns the corrected Gherkin. Lightweight — uses ~300-500 tokens vs ~5k for a full regeneration. Requires BYOK LLM key.

Fetch a Confluence page and return its content as clean Markdown. Accepts a numeric page_id or a full page URL. Optionally lists direct child pages. BYOK — credentials transit in-memory only, never stored.

Returns what the server knows about the current MCP client: clientInfo captured during initialize, User-Agent, and any _meta fields sent with this request. Useful for debugging caller identification.

Give honest usage feedback on an IA-QA MCP tool. Provide a score (1-5) and a comment. Rate low (1-2) if the tool was wrong, irrelevant, or a poor fit; rate high (4-5) only if it genuinely solved your need. Ratings are read by the IA-QA maintainers and are not published, one vote per agent and tool a month. Comments are screened before storing: one carrying a credential is not stored and personal data is replaced by placeholders — describe what happened rather than what a model should do. Skip rating routine successes — we want signal, not praise. Example: rate_tool({ tool_name: "format_json", score: 2, comment: "Tried to pretty-print a JSON5 file, it rejected trailing commas — not usable for my case." })

Validate an email address against RFC 5322 syntax before storing it, sending a transactional email, or adding it to a mailing list. Returns { valid, email } — use this to avoid bounces and malformed data.

Minify a JavaScript snippet, function, class, or module up to 50 KB using Terser. Returns minified code and byte savings. Use when embedding scripts in HTML templates, report payloads, or injecting inline code programmatically.

Decode a JWT (JSON Web Token) and return its header and payload without verifying the signature. Also reports whether the token is expired and the exact expiry date. Use to inspect claims (sub, iss, exp, roles) during debugging or when integrating with an auth provider.

Convert between Unix timestamps (seconds or milliseconds) and ISO-8601 / UTC date strings. Auto-detects epoch vs. millisecond format. Omit input to get the current time. Returns iso, unix_s, unix_ms, utc, date, and time fields.

Parse and validate a URL. Returns decomposed components: protocol, hostname, port, path, query parameters, hash, and origin.

Calculate WCAG 2.1 contrast ratio between two colors. Returns ratio and compliance for AA/AAA normal and large text.

Parse a cron expression into a human-readable schedule description. Supports standard 5-field cron (minute hour day month weekday).

Validate a 5-field cron expression, explain the schedule, and preview the next execution times. Use this to debug cron jobs before they reach production. Returns parsed fields, a human-readable description, and upcoming ISO timestamps.

Look up detailed information about any HTTP status code: class, name, description, cacheability, typical causes, and handling best practices. Covers every code in the IANA HTTP Status Code Registry (1xx-5xx, including 226, 425, 451, 508, 511 and the WebDAV codes) with its defining RFC; anything outside the registry is reported as `registered: false` rather than described. `cacheable` means heuristically cacheable by default per RFC 9110 §15.1.

Parse a raw HTTP headers block into a structured JSON object. Detects multi-value headers, masks Authorization values, and optionally audits for missing security headers (HSTS, CSP, X-Frame-Options, etc.).

Generate a curl command from request parameters. Supports GET/POST/PUT/DELETE, custom headers, JSON body, and form data. Useful for documentation, sharing, and debugging API calls.

Extract TODO, FIXME, HACK, BUG, NOTE, OPTIMIZE, and custom tags from any source code or text. Returns line numbers, tag types, and message text. Essential for technical debt auditing.

Scan code or config files for hardcoded secrets: AWS keys, GitHub tokens, OpenAI/Anthropic API keys, Stripe secrets, JWTs, database connection strings, and generic passwords. Returns findings with severity. Run before every commit.

Count lines of code: total, code lines, comment lines, blank lines, and comment density. Supports JS/TS, Python, Java/C/C++, Ruby, Go, Shell, HTML/XML, and CSS.

Validate a git commit message against the Conventional Commits spec (feat, fix, docs, style, refactor, test, chore, ci, perf, build). Returns compliance score, breaking change detection, and actionable suggestions.

Convert raw byte counts to human-readable sizes in SI (KB=1000) or IEC (KiB=1024) units, or parse size strings back to bytes. Covers B, KB/KiB, MB/MiB, GB/GiB, TB/TiB, PB/PiB.

Compound quality gate for pull requests. Runs three sequential checks: (1) secret detection — scans diff for API keys, tokens, passwords matching 16 regex patterns; (2) bug analysis — heuristic scan for eval(), innerHTML, empty catch, console.log, TODO/FIXME; (3) commit message linting against Conventional Commits spec. Returns gate verdict (PASS/WARN/BLOCK), blockers, and actionable warnings. Use before merging any code change.

Create a temporary webhook endpoint that captures incoming HTTP requests for one hour. Returns the webhook id, public URL, expiration timestamp, and current request count. Use together with webhook_endpoint_requests to inspect captured payloads.

Fetch the requests captured by a webhook created with webhook_endpoint_create. Returns the newest requests first with method, headers, query params, body payload, and timestamps.

Parse a .env file content into a JSON object. Handles quoted values (single and double), inline comments, export prefix, and escaped sequences (\n, \t inside double quotes). Returns all key-value pairs. Use in CI/CD pipelines, agent config loaders, or when processing dotenv files programmatically.

Validate the structure of an OpenAPI 3.x specification (JSON or YAML). Checks required top-level fields (openapi, info.title, info.version, paths), validates each operation (responses, operationId uniqueness), detects undeclared $ref components, and flags missing 2xx responses. Returns a PASS/FAIL verdict, a 0–100 compliance score, and a list of errors and warnings with JSON-pointer locations. Use before publishing an API spec or generating SDK code.

Convert any string into a URL-friendly slug: lowercase, ASCII-normalized (é→e), special characters removed, spaces replaced with hyphens. Use for generating SEO-friendly URL paths, file names, or identifier keys from user-provided titles or labels.

Fetch the latest QA & AI/LLM articles aggregated from curated RSS sources (Google Testing Blog, DEV.to Testing/QA/AI/LLM/Agents, Hugging Face Blog, Simon Willison). Perfect for agents monitoring the QA & AI landscape. Each article carries summary_source — the XML tag the summary was read from, or "none" when the feed publishes titles and links only; an empty summary with summary_source "none" is a property of that feed, not a parse failure.

Analyze a webpage <head> HTML (or full HTML) for GEO (Generative Engine Optimization) signals. Returns a score /60 with per-check results and improvement tips. GEO = optimizing pages for AI-powered search engines (ChatGPT Search, Perplexity, etc.).

Measure response time of one or more HTTP endpoints (GET/POST). Runs N iterations and returns min/max/avg/p95 latency. Useful for API and MCP server benchmarking.

Validate and pretty-print a string that is ALREADY valid JSON. Strict by design — it is a validity gate: valid JSON comes back formatted, anything else is rejected with the exact parse error. It never repairs, completes, or guesses. NOT for: plain text or prose (will fail), JSON embedded in markdown/prose (use extract_json_from_text first), JS objects (JSON.stringify them first), YAML (use yaml_to_json).

Parse a CSV string into a JSON array of objects (or raw arrays). Full RFC 4180: quoted fields may contain the delimiter, embedded newlines (the Excel/Sheets multi-line cell), and doubled quotes. Custom delimiters supported. An unterminated quote is rejected with its position rather than parsed into corrupted rows. Use when processing spreadsheet exports, data imports, or structured text pipelines where the source is CSV. Supports up to 200 KB.

Extract a value from a JSON string using dot-notation path (e.g., "user.address.city", "items.0.name", "meta.tags"). Supports array index access via numeric path segments.

Generate a ready-to-paste <script type="application/ld+json"> snippet for GEO / structured data optimization. Supported types: WebSite, FAQPage, Article, Person, Organization, SoftwareApplication, HowTo.

Validate a JSON value against a JSON Schema (draft-07 subset). Supports type, required, properties, items, enum, const, pattern, format (email/uri/date/date-time/uuid), minimum/maximum, minLength/maxLength, minItems/maxItems, uniqueItems, additionalProperties, anyOf, allOf, oneOf, not, and local $ref (#/definitions/…, #/$defs/…). A schema using a keyword outside that list (patternProperties, if/then, dependencies, a remote $ref…) is never reported valid: the keyword is named in `unevaluated_keywords`. Returns all validation errors with dot-notation paths.

Flatten a nested JSON object to single-level dot-notation keys (e.g. {"a":{"b":1}} → {"a.b":1}), or unflatten dot-notation keys back to a nested object. Supports custom separators.

Convert an XML string to a JSON object. Supports attributes, nested elements, arrays, CDATA, and namespaces. Options: parse numbers, parse booleans, ignore attributes.

Generate realistic mock data from a JSON Schema. Supports all common types (string, number, integer, boolean, array, object, null), format hints (email, date, date-time, uri, uuid), enum, const, and nested schemas. Perfect for testing MCP tools with realistic data.

Transform a JSON array using common operations: pluck (extract specific fields), filter (by field value), sort_by (field), group_by (field), count_by (field), uniq_by (field). Useful for processing MCP tool results and LLM structured outputs.

Convert a JSON array of objects to CSV format. Automatically detects columns from all object keys. Handles quoting and escaping per RFC 4180.

Compute a deep structural diff between two JSON values. Returns added, removed, and changed keys with dot-notation paths. Like git diff but for JSON objects — perfect for API response regression testing. Arrays are not compared blindly by position: the same elements in a different order collapse to a single "reordered" change, and an array of records sharing a stable identity field (id, uuid, key, name…) is matched by that field, so paths read [id=42] and a moved record is not reported as N rewrites.

Deep merge two JSON objects. Supports three array strategies: replace (default), concat, or unique (dedup concat). Nested objects are recursively merged — override takes precedence for primitives.

Convert a JSON object to clean, human-readable YAML. Handles nested objects, arrays, multiline strings, and special characters. No external dependencies.

Parse a YAML string and return the equivalent JSON value. The reverse of json_to_yaml. Supports nested objects, arrays, anchors, aliases, multi-document streams, and all scalar types. Use when processing config files, CI/CD pipeline definitions, or OpenAPI specs authored in YAML.

Infer a JSON Schema (draft-07) from a sample JSON value. Detects types, required fields, array item shapes, nested objects, and common string formats (email, uri, date, date-time, uuid). Returns a ready-to-use schema compatible with json_schema_validate. Use when you have a sample API response or LLM output and want to auto-generate a validation schema for CI/CD testing.

Convert a JSON array of objects into a Markdown table. Automatically detects columns, aligns headers, and fills missing keys with empty cells. Use when an agent needs to present structured data — tool results, model comparisons, test reports — as a readable table in a response or document.

Compute comprehensive statistics for any text: character count (with and without spaces), word count, line count, sentence count, paragraph count, and estimated reading time in minutes. Sentence counting is abbreviation-aware — titles (Mr., Dr.), acronyms (U.S., i.e., p.m.), initials, decimals, URLs and emails do not end a sentence, and a text with no terminal punctuation still counts as one. Use for validating form field lengths, evaluating LLM output verbosity, or content auditing.

Test a regular expression pattern against an input string and return all matches with their index positions and named capture groups. Use for validating user inputs, extracting structured data from text, or debugging regex patterns. Supports flags g, i, m, s, u, y. The match runs in an isolated thread with a 500 ms budget: a pattern that blows up (catastrophic backtracking, e.g. "(a+)+$") comes back as redos_detected:true — a real ReDoS verdict on your pattern — instead of hanging.

Generate Lorem Ipsum placeholder text for UI mockups, design prototypes, or test data population. Configurable paragraphs (1–10), sentences per paragraph (1–20), and approximate words per sentence (3–30).

Strip all Markdown formatting (headers, bold, italic, code fences, links, lists) from text and return clean plain text. Run this before injecting scraped documentation, README files, or user content into an LLM prompt to eliminate redundant markup tokens and reduce cost.

Escape HTML special characters (&, <, >, ", ') to their safe HTML entities. ALWAYS call this before inserting any user-provided or LLM-generated content into an HTML template to prevent cross-site scripting (XSS) attacks.

Convert HTML entities (&amp;, &lt;, &gt;, &quot;, &#x27;, and numeric &#NNN;) back to plain characters. Use when processing HTML-encoded text from APIs, email content, or legacy database fields before passing to an LLM or displaying to users.

Convert a string between naming conventions: camelCase, PascalCase, snake_case, kebab-case, UPPER_SNAKE_CASE, dot.case, Title Case. Essential for code generation and refactoring.

Sort, deduplicate, reverse, or filter lines of text. Useful for cleaning import lists, dependencies, log files, and config entries.

Convert HTML to clean Markdown. Strips scripts, styles, nav, ads, and comments. Converts headings, lists, links, images, code blocks. Ideal for preparing web content as LLM context.

Calculate readability scores: Flesch Reading Ease, Flesch-Kincaid Grade Level, Coleman-Liau Index, and Automated Readability Index. Useful for evaluating LLM output quality.

Normalize whitespace: trim trailing spaces, collapse blank lines, normalize line endings (LF/CRLF), convert tabs to spaces. Useful for cleaning code, configs, and text before processing.

Analyze word frequency in text. Returns top N words with counts and percentages. Supports English stopword filtering. Useful for content analysis, keyword extraction, and LLM output analysis.

Extract all URLs, email addresses, and domain names from text. Returns categorized and deduplicated results. Useful for content auditing, link checking, and web scraping validation.

Compute the Levenshtein (edit) distance and normalized similarity ratio between two strings. Supports batch comparison. Useful for fuzzy string matching, deduplication, and test result comparison.

Detect the natural language of a text using n-gram frequency analysis and common word markers. Supports 15 languages: English, French, Spanish, German, Italian, Portuguese, Dutch, Russian, Chinese, Japanese, Korean, Arabic, Polish, Turkish, Swedish.

Generate one or more cryptographically random UUID v4 identifiers. Use this when you need unique IDs for test fixtures, database records, session tokens, or any scenario requiring a guaranteed-unique string. Returns up to 100 UUIDs in one call.

Compute a cryptographic hash of a text string. Use when you need to verify data integrity, generate content fingerprints, hash passwords (prefer SHA-256+), or produce a fixed-length digest of any input. Supports SHA-256 (default), SHA-512, SHA-1, and MD5.

Encode a UTF-8 string to Base64. Use when you need to embed binary data, multi-line text, or special characters safely inside JSON fields, HTTP headers, or data URIs.

Decode a Base64 string back to UTF-8 text. Use for inspecting Base64-encoded API responses, JWT payload claims, config file values, or attachment data.

Percent-encode a string for safe use in URLs. Call this before programmatically building query strings, path segments, or form-encoded bodies to prevent injection and malformed URLs.

Decode a percent-encoded URL string back to plain text. Use when parsing query parameters from raw URLs or when displaying encoded values to users.

Generate a cryptographically secure random password using crypto.randomBytes. Configurable length (4–128), uppercase letters, digits, and symbols. Use when resetting user passwords, seeding test accounts, or generating API secrets.

Convert a color between HEX, RGB, and HSL formats. Use when translating design tokens between CSS notations, verifying color accessibility, or normalizing color values from user input. Accepts #rrggbb, #rgb, rgb(r,g,b), or hsl(h,s%,l%).

Convert numbers between bases: decimal, binary, octal, hexadecimal, or any base 2–36. Auto-detects 0x, 0b, 0o prefixes.

Compute an HMAC signature for a message using a secret key. Supports SHA-256 (default), SHA-512, SHA-1, and MD5. Used for API request signing, webhook verification (GitHub, Stripe, Twilio), and JWT validation.