
Agent JSON Reliability
io.github.toninovo4249-aiPublished by GitHub account toninovo4249-aiv0.1.0Updated Oct 24, 2026
Deterministic inspect, safe repair, and JSON Schema validation for agent JSON. No LLM.
Overview
Deterministically inspects, safely repairs and schema-validates agent-generated JSON without an LLM, refusing ambiguous input.
- What you can do
- Inspect text to determine whether it is JSON and which failures are present. Apply only structural, unambiguous repairs without inventing values. Validate the repaired result against an optional JSON Schema. Return structured diagnostics covering original validity, repair status, final validity, schema result, ambiguity, changes and errors. Refuse truncated or ambiguous input rather than guessing.
- When to use it
- Use it when an assistant or program emits malformed JSON — markdown fences, trailing commas, single quotes, True/False/None — and you need to inspect, repair and schema-validate it before passing it downstream. Suited to reproducible cleanup that should not rely on an LLM guessing.
- Requirements
- Runs locally as a Python process, started with uvx from the PyPI package agent-json-reliability or as a Python module from a checkout. No account, API key or environment variables are declared. Desktop only.
Installation
In SourceWeft
- Open Agent JSON Reliability in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
Agent JSON Reliability
Deterministic JSON repair + JSON Schema validation for AI agents.
No LLM. Safe refusal on ambiguity.
[PyPI version] [Python] [MCP] [License: MIT]
STATUS=BETA 0.1.0 · GitHub · PyPI · Official MCP Registry
Problem: agents emit malformed JSON (fences, trailing commas, single quotes). This tool inspects, safely repairs, and schema-validates that text. No LLM. No invented values.
Machine catalog: https://agent-json-reliability.onrender.com/v1/catalog
Skill: https://agent-json-reliability.onrender.com/skill.md
Quick start
MCP client config (stdio, published package):
AI agents and software frequently emit malformed JSON: markdown fences, trailing commas, single quotes, True/False/None, and extra prose around one object.
Agent JSON Reliability is a deterministic pipeline for that failure mode:
- inspect — diagnose whether the text is JSON and which failures are present
- safe repair — apply only structural, unambiguous fixes
- validate — optional JSON Schema check on the result
- structured diagnostics —
valid_original,repaired,valid_final,schema_valid,unsafe_or_ambiguous,changes,errors
This is not a generic jsonschema wrapper. Missing semantic values are never invented. Ambiguous input is refused.
Primary HTTP endpoint (local/self-hosted): POST /v1/json/reliable
MCP tools (same core functions): reliable_json, validate_json, repair_json, inspect_json
Before / after (safe repair)
Malformed agent output:
Expected result (semantically equivalent valid JSON):
valid_original=falserepaired=truevalid_final=trueunsafe_or_ambiguous=falsejson→{"name":"alice","age":30}
If a schema is supplied and satisfied: schema_valid=true.
Safe refusal (ambiguous)
Truncated input such as {"user": is JSON-intended but not deterministically repairable without inventing keys or values.
Expected:
- HTTP
200(documented safe response) valid_final=falseunsafe_or_ambiguous=trueorrepaired=false
Do not treat a correct refusal as a product failure.
Run locally (no Hunter)
Binds 127.0.0.1:8770 by default. Set PUBLIC_BASE_URL to an HTTPS origin only when you expose the process yourself. Do not commit a temporary tunnel hostname as the canonical URL.
Hosted API (x402)
Public origin: https://agent-json-reliability.onrender.com
Inspect / validate / repair / MCP stay free. POST /v1/json/reliable returns HTTP 402 until an x402 PAYMENT-SIGNATURE settles 3000 atomic USDC (0.003) on Base (eip155:8453) via PayAI.
HTTP examples
See examples/curl.md, examples/python.py, examples/javascript.js.
Remote examples use https://agent-json-reliability.onrender.com. Local default is http://127.0.0.1:8770.
Share URLs (optional, unverified telemetry tags):
/?source=github/?source=mcp-registry/?source=api-directory
MCP (stdio — durable)
Package transport does not depend on a temporary public URL. Prefer uvx (above). From a checkout:
Remote HTTP MCP (POST /mcp) is for a running instance.
Machine discovery
- Hosted origin: https://agent-json-reliability.onrender.com
GET /.well-known/agent-services.jsonGET /.well-known/agent.jsonGET /.well-known/x402GET /openapi.jsonGET /llms.txt·GET /llms-full.txt·GET /AGENTS.md·GET /sitemap.xml·GET /robots.txtPOST /mcpJSON-RPC (tools/list,tools/call)
What this package does not include
No Hunter market database, no private reports, no Windows user paths, no collectors, no wallets in this package. Hosted /v1/json/reliable is x402-paid; local MCP and inspect/validate/repair are free.
Source: README.md at commit 85cb45c
Tools
0Version history
1- v0.1.0LatestSep 21, 2026
Score
40% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts
Declare a license in the repository or the registry entry.