mcp-hooker
Configurable MCP server that converts any OpenAPI specification into MCP tools over HTTP, with support for dynamic reloading and environment variable expansion for secrets.
README
mcp-hooker
Configurable FastMCP server that turns any OpenAPI spec (local or remote, JSON or YAML) into an MCP tool surface over HTTP.
Configuration is YAML via yayaya — layered files, dot-path lookups, and ${ENV_VAR} expansion for secrets.
Project layout
mcp-hooker/
pyproject.toml
config.yaml # default config (safe to commit)
config.local.yaml # optional overlay (gitignored)
Dockerfile
docker-compose.yml
.gitlab-ci.yml
docs/RUNBOOK.md
mcp_hooker/
settings.py # yayaya loader
spec_loader.py # fetch/parse OpenAPI
server.py # reloadable HTTP MCP server
Quick start
cd mcp-hooker
python -m venv .venv
source .venv/bin/activate
pip install -e .
# Edit config.yaml (openapi.spec + api.base_url), then:
mcp-hooker
The server listens on server.host / server.port from config (default 0.0.0.0:8000).
Configuration
Loaded from config.yaml, with optional config.local.yaml merged on top. Override paths with:
export MCP_HOOKER_CONFIG_FILES=config.yaml,config.docker.yaml
Keys
| Path | Description |
|---|---|
server.name |
MCP server display name |
server.host |
Bind address |
server.port |
Bind port |
openapi.spec |
Remote URL or local path to OpenAPI JSON/YAML |
openapi.fetch_timeout |
Timeout when downloading remote specs (seconds) |
api.base_url |
Upstream API base URL (falls back to servers[0].url in the spec) |
api.timeout |
httpx timeout for tool calls |
api.headers |
Extra request headers; values support ${ENV_VAR} |
reload.token |
If set, POST /admin/reload requires Authorization: Bearer <token> |
reload.interval_seconds |
Auto-reload interval (0 = off) |
reload.on_sighup |
Reload on SIGHUP (default true) |
Example api.headers with a secret from the environment:
api:
headers:
Authorization: "Bearer ${API_TOKEN}"
HTTP endpoints
| Endpoint | Method | Purpose |
|---|---|---|
/health |
GET | Liveness + current spec/base URL |
/admin/reload |
POST | Re-read config and OpenAPI spec |
/mcp |
* | Streamable HTTP MCP (FastMCP default) |
Reloading
- HTTP:
curl -X POST http://localhost:8000/admin/reload - Signal:
kill -HUP <pid>whenreload.on_sighupis true - Interval: set
reload.interval_secondsto poll remote specs
Each reload closes the previous httpx client, re-parses config via yayaya, fetches the spec, and rebuilds the FastMCP OpenAPI provider.
Docker
docker build -t mcp-hooker .
docker run --rm -p 8000:8000 -v "$PWD/config.yaml:/app/config.yaml:ro" mcp-hooker
Or with Compose:
docker compose up --build
GitLab CI/CD
.gitlab-ci.yml mirrors zpl-mcp: build on a tagged runner, push $CI_REGISTRY_IMAGE:$CI_COMMIT_SHORT_SHA, and tag latest on the default branch.
Register a runner with tag mcp-hooker (or change the tag in .gitlab-ci.yml to match your fleet).
Cursor / MCP client
Point your client at the streamable HTTP URL, for example:
{
"mcpServers": {
"my-api": {
"url": "http://localhost:8000/mcp"
}
}
}
Development
pip install -e ".[dev]"
ruff check mcp_hooker
Operations
See docs/RUNBOOK.md for architecture, deployment notes, and troubleshooting.
License
MIT
推荐服务器
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。
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。