graphql-mcp 2.1.1

A framework for building Python GraphQL MCP servers.

GraphQL-MCP

Documentation | PyPI | GitHub

---

Instantly expose any GraphQL API as MCP tools for AI agents and LLMs.

GraphQL MCP works with any Python GraphQL library—Strawberry, Ariadne, Graphene, graphql-core, or graphql-api. If you already have a GraphQL API, you can expose it as MCP tools in minutes.

Don't want to run anything? Bridge is a hosted MCP proxy built on graphql-mcp. Paste any public GraphQL URL into your MCP client and every query and mutation shows up as a tool — no code, no deploy.

Features

• Universal Compatibility - Works with any GraphQL library that produces a graphql-core schema
• Automatic Tool Generation - GraphQL queries and mutations become MCP tools instantly
• Remote GraphQL Support - Connect to any existing GraphQL endpoint
• Type-Safe - Preserves GraphQL types and documentation
• Built-in Inspector - Web interface for testing MCP tools
• Multiple Transports - HTTP, SSE, and streamable-HTTP support

Installation

pip install graphql-mcp

Quick Start

With Strawberry (Popular)

Already using Strawberry? Expose it as MCP tools:

import strawberry
from graphql_mcp.server import GraphQLMCP
import uvicorn

@strawberry.type
class Query:
    @strawberry.field
    def hello(self, name: str = "World") -> str:
        return f"Hello, {name}!"

schema = strawberry.Schema(query=Query)

# Expose as MCP tools
server = GraphQLMCP(schema=schema._schema, name="My API")
app = server.http_app()

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8002)

That's it! Your Strawberry GraphQL API is now available as MCP tools.

With Ariadne

Using Ariadne? Same simple integration:

from ariadne import make_executable_schema, QueryType
from graphql_mcp.server import GraphQLMCP

type_defs = """
    type Query {
        hello(name: String = "World"): String!
    }
"""

query = QueryType()

@query.field("hello")
def resolve_hello(_, info, name="World"):
    return f"Hello, {name}!"

schema = make_executable_schema(type_defs, query)

# Expose as MCP tools
server = GraphQLMCP(schema=schema, name="My API")
app = server.http_app()

With Graphene

Graphene user? Works seamlessly:

import graphene
from graphql_mcp.server import GraphQLMCP

class Query(graphene.ObjectType):
    hello = graphene.String(name=graphene.String(default_value="World"))

    def resolve_hello(self, info, name):
        return f"Hello, {name}!"

schema = graphene.Schema(query=Query)

# Expose as MCP tools
server = GraphQLMCP(schema=schema.graphql_schema, name="My API")
app = server.http_app()

With graphql-api (Recommended)

For new projects, we recommend graphql-api for its decorator-based approach:

from graphql_api import GraphQLAPI, field
from graphql_mcp.server import GraphQLMCP

class API:
    @field
    def hello(self, name: str = "World") -> str:
        return f"Hello, {name}!"

api = GraphQLAPI(root_type=API)
server = GraphQLMCP.from_api(api)
app = server.http_app()

Remote GraphQL APIs

Already have a GraphQL API running? Connect to it directly:

from graphql_mcp.server import GraphQLMCP

# Connect to any GraphQL endpoint
server = GraphQLMCP.from_remote_url(
    url="https://api.github.com/graphql",
    bearer_token="your_token",
    name="GitHub API"
)

app = server.http_app()

Works with:

• GitHub GraphQL API
• Shopify GraphQL API
• Hasura
• Any public or private GraphQL endpoint

Examples

The examples/ directory contains four runnable servers:

Example	What it demonstrates	Live
Hello World	Minimal MCP server with a single query	Try it
Task Manager	CRUD, enums, mutations, UUID/datetime scalars	Try it
Nested API	Nested tools, @mcp directive, Pydantic models, async resolvers	Try it
Remote API	Wrap a public GraphQL API via from_remote_url()	Try it

See the examples documentation for detailed walkthroughs.

Documentation

Visit the official documentation for comprehensive guides, examples, and API reference.

Key Topics

• Getting Started - Quick introduction and basic usage
• Configuration - Configure your MCP server
• Remote APIs - Connect to existing GraphQL APIs
• Examples - Real-world usage examples
• API Reference - Complete API documentation

How It Works

GraphQL MCP automatically:

• Analyzes your GraphQL schema
• Generates MCP tools from queries and mutations
• Maps GraphQL types to MCP tool schemas
• Converts naming to snake_case (e.g., addBook → add_book)
• Preserves all documentation and type information
• Supports the @mcp directive to customize how fields and arguments are exposed to MCP — see below

Customizing MCP Exposure with @mcp

A unified @mcp directive lets schema authors override how each field or argument surfaces as an MCP tool. It accepts three optional arguments:

Arg	Type	Effect
name	String	Override the MCP tool/argument name (replaces the default snake_case derivation).
description	String	Override the MCP description (replaces the GraphQL field/argument description).
hidden	Boolean	When true, skip the field or argument from MCP registration entirely.

Valid on FIELD_DEFINITION and ARGUMENT_DEFINITION.

SDL example

directive @mcp(
  name: String
  description: String
  hidden: Boolean
) on FIELD_DEFINITION | ARGUMENT_DEFINITION

type Query {
  getUserById(
    userId: ID! @mcp(name: "id", description: "User UUID")
    debugToken: String @mcp(hidden: true)
  ): User @mcp(name: "fetch_user", description: "Fetch a user by ID.")

  internalMetrics: Metrics @mcp(hidden: true)
}

Python (graphql-api) example

from typing import Annotated
from graphql_api import GraphQLAPI, field
from graphql_mcp import GraphQLMCP, mcp

class API:
    @field
    @mcp(name="fetch_user", description="Fetch a user by ID.")
    def get_user_by_id(
        self,
        user_id: Annotated[str, mcp(name="id", description="User UUID")],
        debug_token: Annotated[str, mcp(hidden=True)] = "",
    ) -> str:
        return f"user:{user_id}"

    @field
    @mcp(hidden=True)
    def internal_metrics(self) -> str:
        return "secret"

api = GraphQLAPI(root_type=API, directives=[mcp])
server = GraphQLMCP.from_api(api)

When an argument is renamed, the MCP tool exposes the new name but the outbound GraphQL query still uses the original argument name — translation happens automatically. Two fields renamed to the same MCP name will raise a ValueError at registration time.

Migrating from @mcpHidden: the previous mcp_hidden directive has been removed. Replace @mcpHidden with @mcp(hidden: true) and the mcp_hidden Python export with mcp(hidden=True).

MCP Inspector

Built-in web interface for testing and debugging MCP tools:

The inspector is enabled by default — visit /graphql in your browser. See the MCP Inspector documentation for details.

Compatibility

GraphQL MCP works with any Python GraphQL library that produces a graphql-core schema:

• Strawberry - Modern, type-hint based GraphQL
• Ariadne - Schema-first GraphQL
• Graphene - Code-first GraphQL
• graphql-api - Decorator-based GraphQL (recommended)
• graphql-core - Reference implementation
• Any GraphQL library using graphql-core schemas

Ecosystem Integration

• graphql-api - Recommended for building new GraphQL APIs
• graphql-db - For database-backed GraphQL APIs
• graphql-http - For HTTP serving alongside MCP

Configuration

# Full configuration example
server = GraphQLMCP(
    schema=your_schema,
    name="My API",
    graphql_http=False,         # Disable GraphQL HTTP endpoint (enabled by default)
    allow_mutations=True,       # Allow mutation tools (default)
)

# Serve with custom configuration
app = server.http_app(
    transport="streamable-http",  # or "http" (default) or "sse"
    stateless_http=True,         # Don't maintain client state
)

See the documentation for advanced configuration, authentication, and deployment guides.

License

MIT License - see LICENSE file for details.