Opentargets

io.github.nickzren由 GitHub 帳號 nickzren 發布v0.7.0更新於 Oct 24, 2026

Read-only Open Targets: targets, diseases, drugs and evidence by name or ID. No API key needed.

已驗證STDIO僅桌面資料與分析AI 與機器學習

概覽

AI 產生的概覽

面向 Open Targets 平台的唯讀 MCP 伺服器,讓助理依名稱或 ID 查詢標的、疾病、藥物、變異及相關證據。

可以做什麼
依名稱或 ID 查詢標的、疾病與藥物,包括藥物的已知標靶藥物與不良事件、疾病的相關標的;名稱有歧義時列出候選供選擇。取得標的-疾病證據、變異詳情與可信集,並在可用時附上資料來源 ID 與 PubMed 文獻引用。搜尋實體並在基因符號、rsID 以及 Ensembl、MONDO、ChEMBL ID 之間對應識別碼。為某種疾病建議藥物重定位候選。對精選工具未涵蓋的內容執行原始 GraphQL 查詢並檢視 schema,還可記錄所用的 API 與資料版本。
適用情境
適合生物醫學或藥物研發類問題,例如某個基因有哪些研發中的藥物、哪些基因與某種疾病相關、某基因與疾病之間有何證據,也適合在不手寫 GraphQL 的情況下探索藥物重定位構想。
執行需求
以 PyPI 套件形式透過 uvx 在本機以 stdio 執行,需要安裝 uv。無需 Open Targets API 金鑰。需要能連線至 Open Targets GraphQL API 的網路;代理設定從 HTTPS_PROXY、HTTP_PROXY 與 NO_PROXY 讀取。可選的環境變數或命令列參數可調整傳輸方式、主機、連接埠、API 位址與速率限制。
安裝前請注意
唯讀:僅查詢 Open Targets 平台,不會向其寫入資料。查詢內容與你查找的實體都會傳送到外部 Open Targets API。臨床報告計數統計的是報告紀錄數而非唯一試驗數,依賴前請先核實。若改用 HTTP 傳輸,預設主機會綁定所有網路介面,會把伺服器暴露到本機之外。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Opentargets,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

Open Targets MCP Server

[CI] [PyPI] [Python] [MCP Registry] [License: MIT]

A read-only MCP server for exploring targets, diseases, drugs and genetic evidence through the Open Targets Platform. Use names or IDs, retrieve supporting evidence, and explore drug-repurposing candidates. No Open Targets API key required.

Install

Requires uv.

Claude Code

bash
claude mcp add opentargets -- uvx opentargets-mcp

Claude Desktop and Cursor: add this to claude_desktop_config.json (Settings → Developer → Edit Config) or ~/.cursor/mcp.json, then restart:

json
{  "mcpServers": {    "opentargets": {      "command": "uvx",      "args": ["opentargets-mcp"]    }  }}

VS Code

bash
code --add-mcp '{"name":"opentargets","command":"uvx","args":["opentargets-mcp"]}'

Codex

bash
codex mcp add opentargets -- uvx opentargets-mcp

Any other stdio MCP client works the same way: command uvx, argument opentargets-mcp.

Tips:

  • If the client reports spawn uvx ENOENT, use the full path from which uvx.
  • Use opentargets-mcp@latest as the argument to pick up new releases.
  • On Apple Silicon, a cryptography build error can indicate uv picked an Intel (Rosetta) Python. uv python find is a first clue; cached tool environments and desktop clients may use a different interpreter.

What you can ask

  • "Which drugs target EGFR, and how far along are they?"
  • "Which genes are most strongly associated with asthma?"
  • "Show the evidence linking BRAF to melanoma."
  • "Suggest drug repurposing candidates for type 2 diabetes."

Use names or IDs: gene symbols, disease and drug names, rsIDs, or Ensembl, MONDO and ChEMBL IDs. If a name matches several entities, the tool lists the candidates so you can pick one.

Tools

68 tools. Run uvx opentargets-mcp --list-tools for the full list.

AreaExamples
Targetsget_target_info, get_target_known_drugs
Diseasesget_disease_info, get_disease_associated_targets
Drugsget_drug_info, get_drug_adverse_events
Evidenceget_target_disease_evidence
Variants and studiesget_variant_info, get_credible_sets
Search and lookupsearch_entities, map_ids
Workflowget_drug_repurposing_candidates
Raw GraphQLgraphql_query, graphql_schema

Many tools accept fields (dot-paths such as ["target.approvedSymbol"]) to return only what you need. For anything the curated tools don't cover, use graphql_query.

Evidence results include datasource IDs and PubMed references where available. Use get_api_metadata to record the Open Targets API and data release with your analysis. clinicalReportCount counts report records, not unique trials; use graphql_query for full reports.

How it works

mermaid
flowchart LR    A(["AI agent<br/>Claude · Codex · VS Code · Cursor"])
    subgraph S["opentargets-mcp"]        direction TB        S1["1. Resolve names to IDs"]        S2["2. Run read-only queries"]        S3["3. Return structured JSON"]        S1 --> S2 --> S3    end
    subgraph O["Open Targets Platform"]        direction TB        API["GraphQL API"]        DB[("ChEMBL · GWAS Catalog<br/>ClinVar · Europe PMC · more")]        API --- DB    end
    A <-->|MCP| S    S <-->|GraphQL| O
    classDef client fill:#e8f0fe,stroke:#4285f4,color:#1f2328    classDef step fill:#ffffff,stroke:#34a853,color:#1f2328    classDef data fill:#ffffff,stroke:#f9ab00,color:#1f2328    class A client    class S1,S2,S3 step    class API,DB data    style S fill:#e6f4ea,stroke:#34a853,color:#1f2328    style O fill:#fef7e0,stroke:#f9ab00,color:#1f2328

Ask in plain language; the server turns names into Open Targets IDs, runs read-only queries and returns structured results. No API key needed.

Configuration

The defaults work for local use. To change them, set environment variables or pass flags (uvx opentargets-mcp --help):

VariableFlagDefault
MCP_TRANSPORT--transportstdio (also http, sse)
FASTMCP_SERVER_HOST--host0.0.0.0
FASTMCP_SERVER_PORT--port8000
OPEN_TARGETS_API_URL--apihttps://api.platform.opentargets.org/api/v4/graphql
OPEN_TARGETS_RATE_LIMIT_RPS--rate-limit-rps0 (off); limits incoming MCP requests
OPEN_TARGETS_RATE_LIMIT_BURST--rate-limit-burst20

Proxies are read from HTTPS_PROXY, HTTP_PROXY and NO_PROXY.

Self-host

From source:

bash
git clone https://github.com/nickzren/opentargets-mcpcd opentargets-mcpuv run opentargets-mcp

With Docker (HTTP on port 8000), from the clone:

bash
docker compose up -d --build

Connect your MCP client to http://localhost:8000/mcp.

Development

bash
uv sync --extra devuv run ruff check src testsuv run pytest -m "not live"   # offlineuv run pytest -m live         # calls the Open Targets API

Example scripts are in examples/.

License

MIT

來源:README.md,提交 e0641af

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包,因此其工具未知。

版本歷史

4
  1. v0.7.0最新Oct 9, 2026
  2. v0.6.1Oct 9, 2026
  3. v0.6.0Sep 21, 2026
  4. v0.2.0Sep 16, 2026

評分

已滿足適用檢查項的 50%

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。

  • 近 30 天內可連通25 分

    請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。

  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分