io.github.srikarjy由 GitHub 帳號 srikarjy 發布v0.3.0更新於 Oct 24, 2026
Auditable MCP server for PubMed, Europe PMC, ClinicalTrials.gov, and bioRxiv/medRxiv queries
可稽核的科研證據伺服器：檢索 PubMed、Europe PMC、ClinicalTrials.gov、UniProt、Open Targets 與 bioRxiv/medRxiv，並以雜湊鏈記錄每次檢索。
Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/io.github.srikarjy%2Fbiolab-mcp-server/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。把它新增到你客戶端的 mcpServers 設定中。
{
"mcpServers": {
"biolab": {
"type": "http",
"url": "https://srikarjy025-biolab-mcp.hf.space/mcp"
}
}
}[Python Version] [Go Version] [MCP] [License] [PyPI] [Docker]
"AI agents querying biological databases leave no audit trail. Six months later, nobody can answer: what exact query returned this result, when, and was that paper peer-reviewed at the time? Biolab solves that."
A local-first scientific evidence workspace with Python and Go clients. It sits between scientists or AI agents and PubMed, Europe PMC, ClinicalTrials.gov, UniProt, Open Targets, and bioRxiv/medRxiv. Every retrieval preserves its source response and returns a retrieval_id; projects, reviewed collections, evidence watches, and claim links build a traceable workflow on top.
New to MCP? It's an open standard that lets an AI assistant — Codex, ChatGPT, Claude, Cursor, and others — call external tools during a conversation. Add Biolab and the assistant can retrieve, preserve, organize, monitor, and cite scientific evidence while keeping every result inspectable later.
A hosted instance is available at https://srikarjy025-biolab-mcp.hf.space/mcp. Point your client at it and you're done. Hosted capabilities follow the version currently deployed there; run locally for the complete 0.4 workspace in this repository.
Claude Code:
Claude Desktop / Cursor — add this to your MCP config file:
(Add "headers": {"Authorization": "Bearer <your key>"} alongside "url" once you have a key — see the rate-limit note below.)
The 0.4 server exposes unified search, source-specific search, standard search/fetch, retrieval inspection, projects, collections, watches, claims, and reports. Every retrieval is written to a hash-chained audit trail you can inspect later (see Audit Trail Schema below).
For cross-source research, search_evidence searches PubMed, Europe PMC,
ClinicalTrials.gov, UniProt, and Open Targets through one call while preserving
explicit results and errors for each source.
Also listed on the official MCP Registry and Smithery if you'd rather discover/install it from there.
A note on rate limits. The hosted server is shared and stays open — no signup required for casual use — but callers with no API key share one small, low-throughput budget (1 req/s to PubMed) so no single anonymous user can starve everyone else. If you're doing more than a handful of queries, ask for a key (below) and you get your own isolated, higher budget instead.
Getting a key:
Add that header in your client's MCP config (Claude Code: claude mcp add --transport http biolab <url> --header "Authorization: Bearer <key>"). Keys are issued with biolab keys create <label> — see Managing API Keys below if you're running your own instance; otherwise ask the maintainer for one.
Want to run your own copy instead (local dev, your own storage, self-hosting)? Keep reading.
A drug discovery team uses an AI agent to research gene targets. The agent queries PubMed 200 times over three days and surfaces a paper claiming gene X is upregulated in pancreatic cancer. A scientist makes a decision based on that. Six months later, during FDA submission:
Without Biolab, nobody can answer any of those questions. The retrieval is invisible.
Biolab is an interception and logging layer, not a retrieval layer. It doesn't interpret evidence, rank it, or summarize it — it records what happened, verbatim, so an agent's claim can always be traced back to an unforgeable original.
The agent gets the paper it asked for. Biolab gets a permanent, queryable, tamper-evident record of exactly what happened.
All sources share a single audit database (SQLite locally, or Turso — a hosted, SQLite-compatible database — in production) with one source-agnostic schema.
You don't need to know Python or Go to get this running locally — just follow these steps in order. All commands are run in a terminal.
python3 --version. Get it from python.org if you don't have it.git --version.That's genuinely it for the Python path — no database server to install, no API keys required (PubMed works anonymously, just at a lower rate limit).
This searches PubMed for real, stores every result in a local biolab.db file (created automatically, no setup needed), and prints back the retrieval_id for each paper — the same ID an AI agent would get back over MCP.
Run pytest -m live -v separately when network access and local socket binding
are available. Those tests exercise real PubMed, Europe PMC, ClinicalTrials.gov,
and bioRxiv services; the default command stays deterministic and offline.
This starts an HTTP server on http://localhost:8000/mcp — point Claude Desktop, Claude Code, or Cursor at that URL exactly like in Use It Now, just with localhost:8000 instead of the hosted URL.
If you'd rather not install Python locally at all:
(Storage defaults to an ephemeral file inside the container unless you set TURSO_DATABASE_URL/TURSO_AUTH_TOKEN — see Environment Variables below.)
The dashboard binds to 127.0.0.1 by default. REST endpoints cover evidence,
projects, collections, watches, claims, and reports. Discover them at
/openapi.json; use /health, /metrics, and /v1/verify for operations.
Run biolab web --require-auth with keys created by biolab keys create, or use
BIOLAB_REQUIRE_AUTH=true on the unified MCP server, before exposing it remotely.
A ready-to-copy BioClaw skill is included at
integrations/bioclaw/biolab-evidence. It connects BioClaw's containerized
research assistant to the local REST API and supports cross-source search,
retrieval inspection, and chain verification without extra Python packages.
python -m biolab.server.integrations/codex/config.toml.example into your Codex configuration and
keep tool approval on prompt while developing.plugins/biolab/ (plugin.json,
mcp.json, and its evidence-workspace skill). Change the remote URL in
mcp.json to your HTTPS deployment before publishing or submitting it.search and fetch knowledge tools plus the
richer Biolab tools. Tool annotations disclose read-only, write, external-data,
and idempotency behavior to clients.ChatGPT/Codex can therefore search and preserve evidence, create project state,
run watches, and trace claims. The assistant proposes relationships; a scientist
must use human_reviewed when a relationship has actually been reviewed.
One python -m biolab.server process serves MCP at /mcp, the dashboard at /,
and REST under /v1. The important routes are:
The zero-cost Caddy/Docker example in deploy/gateway/ adds a reverse-proxy
boundary and persistent volume. It binds only to localhost by default. Add a real
domain, HTTPS, backups, and organization-specific authorization before public use.
The server stays open to unauthenticated callers by design — but they all share one small, low-throughput rate-limit budget (see Use It Now). Issuing someone a key gives them their own isolated, higher budget instead. This doesn't gate access — it's purely a fairness mechanism so one caller can't starve everyone else's share of PubMed's real rate limit.
The caller sends the key back as Authorization: Bearer <key>. By default a
missing header uses the anonymous tier while an invalid/revoked key returns 401.
Set BIOLAB_REQUIRE_AUTH=true to reject requests without a key as well.
All optional — the server runs with sensible defaults if you set none of these.
Key properties:
retrieval_log.verify_chain(conn) to check the whole log; it returns exactly which row broke, if any.hash_version 2); legacy v1 rows still verify.biolab anchor out.json [--chain retrievals|executions] writes an HMAC-signed checkpoint of a chain head (BIOLAB_ANCHOR_KEY). Keep it outside the database host; biolab verify-anchor out.json then detects tail truncation and full rewrites that the chain alone cannot see.biolab run-tool <tool> -i '{...}' runs a registered tool (sequence_stats; mmseqs2_search when mmseqs is on PATH) and seals an immutable ExecutionRecord with content-addressed artifacts. biolab executions {list,show,verify,replay,artifact} inspects, verifies and replays them. If the record cannot be sealed, no result is returned.go-biolab/) writes the same v2 chain and has verify, anchor and verify-anchor; hashes are byte-compatible with Python and pinned by shared test vectors. The Go port does not implement the execution runtime.Design principles:
retrieval_id is worse than errorStudent-cost path: local SQLite, the CLI, dashboard, MCP server, BioClaw client, and Caddy gateway require no paid service. Free hosted tiers may also work, but their quotas and persistence policies can change; verify them before relying on a public deployment.
MIT — see LICENSE
Srikar Jy — [email protected]
來源：README.md，提交 8d3f5fe
Search several scientific sources in one explicitly audited workflow. Args: query: scientific evidence query sent to each selected source agent_id: calling agent or workflow identity sources: any of pubmed, europepmc, clinicaltrials, uniprot, opentargets max_results: maximum results to preserve per source (1-50)
query必填sourcesagent_id必填max_resultsSearch all evidence sources for ChatGPT/Codex knowledge compatibility.
query必填Fetch one preserved evidence record by retrieval ID.
id必填Create a persistent scientist research project.
name必填descriptionList scientist research projects.
無需輸入參數。
Create a reviewed evidence collection in a project.
name必填project_id必填descriptionList evidence collections, optionally within one project.
project_idAdd or update a retrieval in a scientist-reviewed evidence collection.
noterelevanceretrieval_id必填collection_id必填Return a collection and its reviewed evidence items.
collection_id必填Create a saved query for explicit evidence-change monitoring.
query必填source必填agent_id必填project_id必填max_resultsList saved evidence watches, optionally within one project.
project_idRun one evidence watch and record added, removed, or revised evidence.
watch_id必填List change events previously detected by an evidence watch.
watch_id必填Run all active watches and return explicit per-watch failures.
無需輸入參數。
Create a scientific claim record without asserting that it is true.
claim必填project_id必填Propose or record how one retrieval relates to a claim.
notestance必填claim_id必填assessmentretrieval_id必填Trace a claim to supporting, conflicting, contextual, or insufficient evidence.
claim_id必填Return collections, claims, evidence links, watches, and events for a project.
project_id必填Search PubMed and log every retrieved paper to the audit trail. Args: query: exact search string, sent to PubMed verbatim — no normalization agent_id: which agent is asking, e.g. "aletheia:advocate" max_results: how many papers to retrieve (default 5)
query必填agent_id必填max_resultsSearch Europe PMC and log every retrieved article to the audit trail. Args: query: exact search string, sent to Europe PMC verbatim — no normalization agent_id: which agent is asking, e.g. "aletheia:advocate" max_results: how many articles to retrieve (default 5)
query必填agent_id必填max_resultsSearch ClinicalTrials.gov and log every retrieved study to the audit trail. Args: query: condition/disease search string, sent to ClinicalTrials.gov verbatim agent_id: which agent is asking, e.g. "aletheia:advocate" max_results: how many studies to retrieve (default 5)
query必填agent_id必填max_resultsList bioRxiv/medRxiv preprints by category and log each to the audit trail. No free-text search exists on this API — only category listing (last 30 days). Args: category: category, e.g. "neuroscience", "bioinformatics", or "all" agent_id: which agent is asking, e.g. "aletheia:advocate" max_results: how many preprints to retrieve (default 5) server: "biorxiv" or "medrxiv" (default "biorxiv")
serveragent_id必填category必填max_resultsList runnable tools with their pinned versions, containers and devices.
無需輸入參數。
Run a registered tool and seal an ExecutionRecord (inputs, versions, output hashes). Args: tool: tool name from list_execution_tools inputs: tool inputs (JSON object); stored verbatim so the run can be replayed agent_id: calling agent or workflow identity retrieval_ids: evidence retrievals this run was derived from, for provenance links
tool必填inputs必填agent_id必填retrieval_idsFetch one execution record, including output artifact hashes.
execution_id必填List recent execution records, newest first.
toollimitFetch an output artifact by its sha256 (text only; capped in size).
sha256必填Verify the execution hash chain and that every output artifact is intact.
無需輸入參數。
Re-run a recorded execution and report whether its outputs reproduced. Writes nothing.
execution_id必填Retrieve a full retrieval record by its retrieval_id. Args: retrieval_id: the UUID returned by search_pubmed
retrieval_id必填已滿足適用檢查項的 100%
該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。
claude mcp add --transport http biolab https://srikarjy025-biolab-mcp.hf.space/mcp{ "mcpServers": { "biolab": { "url": "https://srikarjy025-biolab-mcp.hf.space/mcp" } }}Authorization: Bearer <your key>Your AI Agent ↓ MCP tool call (e.g. search_pubmed)Biolab MCP Server ↓ HTTPSource API (PubMed, Europe PMC, ClinicalTrials.gov, bioRxiv/medRxiv) ↓ paperBiolab writes a hash-chained retrieval record to the audit database ↓ paper + retrieval_idBack to your agentgit clone https://github.com/srikarjy/biolab-mcp-server.gitcd biolab-mcp-serverpython3 -m venv .venv # creates an isolated Python environmentsource .venv/bin/activate # on Windows: .venv\Scripts\activatepip install -e ".[dev]" # installs the package + test tools# Add the optional remote Turso driver only if needed: pip install -e ".[turso]"biolab demo --query "BRCA1 pancreatic cancer"pytest -m "not live and not benchmark" -vpython -m biolab.serverdocker build -f space/Dockerfile -t biolab-mcp .docker run -p 8000:8000 biolab-mcppipx install biolab-mcp # or: pip install biolab-mcp# Or the Go binary, no Python required at all:curl -L https://github.com/srikarjy/biolab-mcp-server/releases/latest/download/biolab_darwin_arm64.tar.gz | tar xz./biolab search "BRCA1 pancreatic cancer" --max 3# Search several evidence sources in one audited workflowbiolab evidence "KRAS pancreatic cancer" \ --sources pubmed,europepmc,clinicaltrials,uniprot,opentargets --max 5
# Search PubMedbiolab search "BRCA1 pancreatic cancer" --max 5
# Search Europe PMCbiolab search-europepmc "BRCA1 pancreatic cancer" --max 5
# Search ClinicalTrials.govbiolab search-clinicaltrials "pancreatic cancer" --max 5
# List bioRxiv preprints (no free-text search - API limitation)biolab search-biorxiv neuroscience --max 10biolab search-biorxiv all --server medrxiv --max 10
# Retrieve full audit recordbiolab get <retrieval_id>
# List recent retrievalsbiolab list --source pubmed --limit 10
# Export for analysisbiolab export evidence.jsonl --source clinicaltrials
# Verify the complete tamper-evident audit chainbiolab verify --db biolab.db
# Run demobiolab demo --query "BRCA1 pancreatic cancer"
# Run the offline portfolio demo (no network or scientific claims)biolab portfolio-demo
# Run the real end-to-end scientist workflow against a local serverpython scripts/scientist_workflow_demo.py --url http://127.0.0.1:8000
# Start the private local dashboard and REST APIbiolab web# Open http://127.0.0.1:8000
# Create a project, organize evidence, and trace a claimbiolab projects create "KRAS target review" --description "Pancreatic cancer evidence"biolab collections create <project_id> "Reviewed evidence"biolab collections add <collection_id> <retrieval_id> --relevance relevant --note "Reviewed"biolab claims create <project_id> "KRAS is a therapeutic target"biolab claims link <claim_id> <retrieval_id> --stance supports --assessment human_reviewed
# Monitor a source for added, removed, or revised recordsbiolab watches create <project_id> pubmed "KRAS pancreatic cancer"biolab watches run-all
# Export a complete project reportbiolab projects report <project_id> --output report.md// Search any source{"name": "search_evidence", "arguments": {"query": "KRAS pancreatic cancer", "agent_id": "research:target-validation", "sources": ["pubmed", "europepmc", "clinicaltrials", "uniprot", "opentargets"], "max_results": 5}}{"name": "search_pubmed", "arguments": {"query": "BRCA1 pancreatic cancer", "agent_id": "aletheia:advocate", "max_results": 5}}{"name": "search_europepmc", "arguments": {"query": "BRCA1 pancreatic cancer", "agent_id": "aletheia:advocate", "max_results": 5}}{"name": "search_clinicaltrials", "arguments": {"query": "pancreatic cancer", "agent_id": "aletheia:advocate", "max_results": 5}}{"name": "search_biorxiv", "arguments": {"category": "neuroscience", "agent_id": "aletheia:advocate", "max_results": 5, "server": "biorxiv"}}
// Retrieve full audit record (works for ALL sources){"name": "get_retrieval", "arguments": {"retrieval_id": "uuid-from-search"}}from biolab.pubmed_client import search_and_fetchfrom biolab.retrieval_log import write_retrieval, get_retrievalfrom biolab.db import connect
conn = connect("biolab.db")papers = search_and_fetch("BRCA1 pancreatic cancer", 3)for p in papers: record = write_retrieval(conn, query="...", pmid=p.pmid, ...) print(record.retrieval_id)# Issue a key — the raw key is shown once, save it immediatelybiolab keys create alice
# List issued keys (never shows the raw key — only a hash is stored)biolab keys list
# Revoke all of a label's active keysbiolab keys revoke aliceCREATE TABLE retrievals ( retrieval_id TEXT PRIMARY KEY, -- UUID source TEXT NOT NULL, -- pubmed, europepmc, trials, UniProt, Open Targets, etc. external_id TEXT NOT NULL, -- PMID, NCT ID, DOI, etc. query_text TEXT NOT NULL, -- exact query sent to source retrieved_at TEXT NOT NULL, -- ISO 8601 UTC agent_id TEXT NOT NULL, -- e.g. "aletheia:advocate" source_metadata TEXT NOT NULL, -- JSON: source-specific fields raw_response TEXT NOT NULL, -- verbatim XML/JSON from source snapshot TEXT NOT NULL, -- JSON: structured fields (title, abstract, authors, journal, DOI, pub types, MeSH/conditions) response_hash TEXT NOT NULL, -- SHA-256(prev_hash + raw_response + retrieval_id + retrieved_at) prev_hash TEXT NOT NULL -- response_hash of the previous row — makes this a hash chain);biolab/├── cli.py # Search plus scientist workspace commands├── server.py # Unified MCP + REST + dashboard process├── evidence.py # Shared multi-source orchestration and audit persistence├── workspace.py # Projects, collections, watches, claims, reports├── web.py # Local dashboard + versioned REST API├── db.py # Connection + schema (local SQLite or remote Turso)├── models.py # RetrievalRecord dataclass├── retrieval_log.py # Only writer + background queue + hash chain├── pubmed_client.py # PubMed E-utilities wrapper + rate limiter├── europepmc_client.py # Europe PMC adapter├── clinicaltrials_client.py # ClinicalTrials.gov adapter├── uniprot_client.py # UniProt adapter├── opentargets_client.py # Open Targets GraphQL adapter├── biorxiv_client.py # bioRxiv/medRxiv adapter└── migrations/ # Schema migration scripts
integrations/├── bioclaw/ # BioClaw evidence skill and REST client└── codex/ # Local MCP configuration template
plugins/biolab/ # Portable Codex/ChatGPT Agent Plugin bundledeploy/gateway/ # Caddy + Docker Compose deployment template
space/ # Files pushed to the hosted Hugging Face Space├── Dockerfile # Python-server-specific image (see repo-root Dockerfile for the Go one)└── README.md # Space config (title, hosting metadata)# Pythonpip install -e ".[dev]"pytest -m "not live and not benchmark" -v
# Gocd go-biolabgo test ./...go build -o biolab ./cmd/cligo build -o biolab-server ./cmd/server