graphql-mcp-server
Automatically generates MCP tools from any GraphQL API by introspecting its schema, supporting queries, mutations, and authentication.
README
GraphQL MCP Server
A Model Context Protocol (MCP) server that dynamically generates tools for any GraphQL API by introspecting its schema.
Features
- Automatic tool generation from GraphQL schema introspection
- Support for all GraphQL queries and mutations
- Type-safe variable handling
- Automatic query generation with smart field selection
- Error handling and validation
- Function exposure control via exposed.yaml
- Bearer token authentication support
Function Exposure Control
The server automatically manages which GraphQL functions are exposed as MCP tools through an exposed.yaml file:
How it works
- First run: Server discovers all GraphQL queries and mutations, creates
exposed.yamlwith all functions set totrue - Subsequent runs: Only functions marked as
trueinexposed.yamlare registered as MCP tools - Dynamic updates: New functions are automatically added to the file with default
truevalue - Cleanup: Functions that no longer exist in the schema are removed from the file
exposed.yaml format
exposed:
queries:
getUser: true
listPosts: false
searchContent: true
mutations:
createUser: true
deleteUser: false
updatePost: true
Managing exposed functions
- Set a function to
falseto disable it (won't be registered as MCP tool) - Set a function to
trueto enable it (will be registered as MCP tool) - The file is automatically updated when the schema changes
Installation
npm install
Configuration
GraphQL Endpoint
Set the GraphQL endpoint URL via environment variable or command line argument:
# Via environment variable
export GRAPHQL_URL="https://api.example.com/graphql"
# Via command line argument
node src/index.js --graphql-url "https://api.example.com/graphql"
GraphQL API Authentication
For GraphQL APIs requiring Bearer token authentication:
# Via environment variable
export GRAPHQL_TOKEN="your-bearer-token"
# Via command line argument
node src/index.js --token "your-bearer-token"
# Both URL and token as arguments
node src/index.js --graphql-url "https://api.example.com/graphql" --token "your-bearer-token"
Usage
Running the server
STDIO Transport (default)
npm start
# or
node src/index.js
HTTP Transport
npm run start:http
# or
node src/index.js --transport http
# Custom port
node src/index.js --transport http --port 8080
Development mode
# STDIO transport
npm run dev
# HTTP transport
npm run dev:http
Command Line Options
node src/index.js [options]
Options:
-t, --transport <type> Transport type: stdio or http (default: stdio)
-p, --port <number> HTTP port (default: 3000)
-u, --graphql-url <url> GraphQL endpoint URL
-T, --token <token> Bearer token for authentication
-q, --query-prefix <str> Prefix for query tools (default: none)
-m, --mutation-prefix <str> Prefix for mutation tools (default: none)
-h, --help Show help message
Integration with Claude Desktop
STDIO Transport
Add to your Claude Desktop configuration:
{
"mcpServers": {
"graphql": {
"command": "node",
"args": ["/path/to/graphql-mcp-server/src/index.js"],
"env": {
"GRAPHQL_URL": "https://api.example.com/graphql",
"GRAPHQL_TOKEN": "your-bearer-token"
}
}
}
}
Alternative using command line arguments:
{
"mcpServers": {
"graphql": {
"command": "node",
"args": [
"/path/to/graphql-mcp-server/src/index.js",
"--graphql-url", "https://api.example.com/graphql",
"--token", "your-bearer-token"
]
}
}
}
HTTP Transport
For HTTP transport, start the server separately:
# Start the server with environment variables
GRAPHQL_URL="https://api.example.com/graphql" GRAPHQL_TOKEN="your-bearer-token" node src/index.js --transport http --port 3000
# Or using command line arguments
node src/index.js --transport http --port 3000 --graphql-url "https://api.example.com/graphql" --token "your-bearer-token"
Then configure your MCP client to connect to the HTTP endpoint at http://localhost:3000/mcp.
MCP Server Authentication (HTTP Transport)
When using HTTP transport, you can secure the MCP server with bearer token authentication:
# Set AUTH_TOKEN environment variable
export AUTH_TOKEN="your-mcp-secret-token"
node src/index.js --transport http
# Or in docker-compose.yml
environment:
- AUTH_TOKEN=your-mcp-secret-token
When AUTH_TOKEN is set, clients can authenticate using either:
-
Authorization header (recommended):
Authorization: Bearer your-mcp-secret-token -
URL query parameter:
http://localhost:3000/mcp?token=your-mcp-secret-token
Authentication behavior:
- If
AUTH_TOKENis set, all requests must provide valid authentication - Requests without valid tokens receive 401 Unauthorized
- If
AUTH_TOKENis not set, the server allows open access (no authentication required) - The server checks the Authorization header first, then falls back to the URL parameter
HTTP API Endpoints
When running with HTTP transport:
- POST /mcp - Main MCP protocol endpoint
- GET /mcp - Returns method not allowed (405)
Example HTTP Usage
# Without authentication (when AUTH_TOKEN is not set)
curl -X POST http://localhost:3000/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}'
# With authentication using header (when AUTH_TOKEN is set)
curl -X POST http://localhost:3000/mcp \
-H "Content-Type: application/json" \
-H "Authorization: Bearer your-mcp-secret-token" \
-d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}'
# With authentication using URL parameter (when AUTH_TOKEN is set)
curl -X POST "http://localhost:3000/mcp?token=your-mcp-secret-token" \
-H "Content-Type: application/json" \
-d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}'
# Call a GraphQL query tool
curl -X POST http://localhost:3000/mcp \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "query_getUser",
"arguments": {
"variables": {
"id": "123"
}
}
}
}'
How it works
- On startup, the server connects to the specified GraphQL endpoint (with optional Bearer token authentication)
- It loads the
exposed.yamlconfiguration file (creates it if it doesn't exist) - It performs an introspection query to fetch the complete schema
- For each query and mutation in the schema, it checks if the function is enabled in
exposed.yaml - Only enabled functions are registered as MCP tools with prefixes:
query_for queries andmutation_for mutations - New functions discovered in the schema are automatically added to
exposed.yamlas enabled - When a tool is called, it constructs the appropriate GraphQL query/mutation and executes it
- Results are returned as formatted JSON
Tool naming convention
- Queries:
query_<fieldName>(e.g.,query_getUser,query_listPosts) - Mutations:
mutation_<fieldName>(e.g.,mutation_createUser,mutation_updatePost)
Environment Variables
| Variable | Description | Required |
|---|---|---|
GRAPHQL_URL |
GraphQL endpoint URL | Yes (if not provided via --graphql-url) |
GRAPHQL_TOKEN |
Bearer token for GraphQL API authentication | No |
AUTH_TOKEN |
Bearer token for MCP server authentication (HTTP transport only) | No |
Files
| File | Description |
|---|---|
exposed.yaml |
Configuration file controlling which GraphQL functions are exposed as MCP tools |
Transport Protocols
The server supports two transport protocols:
STDIO Transport
- Default transport for MCP
- Uses standard input/output for communication
- Suitable for direct integration with MCP clients
- Lower overhead, faster communication
HTTP Transport
- StreamableHTTP transport using MCP SDK
- JSON-RPC over HTTP
- Express.js server for HTTP handling
- Suitable for web applications and remote access
- Standard MCP protocol compliance
Example
For a GraphQL schema with:
type Query {
getUser(id: ID!): User
listPosts(limit: Int): [Post]
}
type Mutation {
createUser(input: CreateUserInput!): User
}
The MCP server will generate tools:
query_getUser- with requiredidparameterquery_listPosts- with optionallimitparametermutation_createUser- with requiredinputparameter
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。