genefoundry

genefoundry

Federates 13 gene-related MCP backends (gnomAD, GTEx, etc.) behind a single Streamable HTTP endpoint with collision-free namespacing and search-based tool discovery.

Category
访问服务器

README

GeneFoundry Router

genefoundry-router — a thin FastMCP 3.x aggregator that federates the GeneFoundry *-link MCP fleet behind a single Streamable-HTTP endpoint (deployed MCP server name: genefoundry). A host (Claude, Cursor, Gemini, …) adds one server and transparently gets every backend (gnomAD, GTEx, HGNC, MGI, UniProt, ClinGen, GenCC, LitVar, STRING, AutoPVS1, SpliceAI, GeneReviews, PubTator) with collision-free namespacing, normalized tool names, and search-based tool discovery.

Research use only. Not for clinical decision support. Mirrors the disclaimers of the underlying -link backends.

Core Purpose

The fleet is ~13 FastMCP -link servers (~189 tools). Exposing all of them to a model at once is unworkable. The router federates them behind one endpoint, namespaces every tool as <token>_<tool> (e.g. gnomad_get_variant_details) so names never collide, and presents a small search surface so the model is never shown the full catalog at once.

Key Features

  • One endpoint federating all -link backends; config-driven registry (servers.yaml).
  • Collision-free namespacing (<token>_<tool>), with a 64-char MCP-name guard.
  • Tool-overload control via FastMCP BM25SearchTransform — exposes search_tools + call_tool + a few pinned essentials instead of ~189 raw tools.
  • Pluggable, OAuth-ready auth (none | jwt | oauth) on the router endpoint.
  • Per-backend normalization transforms (e.g. strip a self-prefix) until source repos adopt Tool-Naming Standard v1.
  • Observability: /health (per-backend reachability), /metrics (Prometheus), structlog JSON logs with correlation IDs.

Quick Start

# 1. Install (Python 3.12+, uv)
uv sync --group dev

# 2. Configure: copy the template and fill in backend URLs
cp .env.example .env        # then edit GF_*_URL values

# 3. Run the router over Streamable HTTP
uv run genefoundry-router run --host 127.0.0.1 --port 8000

# 4. Verify
curl -s localhost:8000/health | python -m json.tool

Add the server to your MCP host using the /mcp URL (e.g. http://localhost:8000/mcp).

Configuration

Structure lives in committed servers.yaml; secrets/URLs live in gitignored .env (copy .env.example). Key environment variables (prefix GF_):

Variable Default Purpose
GF_HOST / GF_PORT 127.0.0.1 / 8000 Bind address
GF_MCP_PATH /mcp MCP mount path
GF_SERVERS_FILE servers.yaml Backend registry
GF_SEARCH_MAX_RESULTS 5 BM25 search_tools result cap
GF_POLL_INTERVAL 0 Polling re-list interval (s); 0 disables
GF_AUTH_MODE none none | jwt | oauth
GF_ALLOWED_ORIGINS (empty) CSV Origin allowlist (DNS-rebinding defense)
GF_PUBLIC_BASE_URL (unset) Public URL behind the proxy (OAuth resource URI)
GF_JWT_ISSUER / GF_JWT_JWKS_URL / GF_JWT_AUDIENCE (unset) JWT verification
GF_<NAME>_URL (unset) Per-backend /mcp URL (e.g. GF_GNOMAD_URL)

A backend with a missing/unset URL is skipped with a warning (the router still starts).

CLI

genefoundry-router run        --host 0.0.0.0 --port 8000   # serve over Streamable HTTP
genefoundry-router validate                                # check servers.yaml + env, report missing URLs
genefoundry-router list-tools [--namespace gnomad]         # enumerate federated tools, flag >64-char names
genefoundry-router doctor     [--strict-naming]            # ping backends; optionally audit leaf names vs Standard v1

MCP Integration & Tool Discovery

The router speaks Streamable HTTP at /mcp. Instead of listing ~189 tools, it exposes a synthetic search surface: search_tools (relevance search over the federated catalog) and call_tool (invoke a discovered tool), plus a few pinned always_visible essentials. Originals remain callable but hidden from the default listing.

This server-side search_tools/call_tool surface is client-agnostic and independent from Anthropic's API-level tool-search. The federated names are also valid for Gemini Remote MCP (snake_case, [a-z0-9_], ≤64 chars, no dots/dashes) — the gateway already emits these.

Architecture

host → Streamable HTTP + auth → genefoundry-router (FastMCP "genefoundry")
         • MultiAuth (none|jwt|oauth)   • BM25SearchTransform (search_tools/call_tool)
         • per-backend ToolTransform    • mount(create_proxy(url), namespace=token)
         • /health, /metrics, structlog
       → 13 remote -link MCP backends (metadata-cached proxies)

Security

  • Origin validation (GF_ALLOWED_ORIGINS): per MCP transport spec, a request that sends a disallowed Origin is rejected with 403 (DNS-rebinding defense). Requests with no Origin (non-browser MCP clients) pass through.
  • Auth modes: GF_AUTH_MODE=none is local/PoC only; use jwt or oauth for public deployments. In jwt/oauth mode the router serves MCP Protected-Resource-Metadata and returns 401 + WWW-Authenticate to unauthenticated callers.
  • No token passthrough: the gateway authenticates the caller at the edge and never forwards the caller's token to the 13 backends (confused-deputy defense).

Deployment

Docker image + compose overlays under docker/ (base / prod / dev / npm), mirroring the fleet. Runs behind nginx-proxy-manager; set GF_PUBLIC_BASE_URL and ensure forwarded headers (X-Forwarded-Proto/-Host) reach the app so generated URLs use the public host.

make docker-build      # build the image
make docker-up         # start the dev stack

Status caveats

  • hgnc is enabled: false until its live deployment is fixed (currently serves the mgi-link binary).
  • pubtator carries a strip_prefix: "pubtator_" transform until pubtator-link drops its self-prefix at the source.

See docs/specs/2026-06-13-genefoundry-router-design.md and docs/plans/2026-06-13-genefoundry-router-implementation.md.

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选