Graphql

graphql-to-mcp

[npm version] [npm downloads] [License: MIT]

Turn any GraphQL API into MCP tools — zero config, zero code.

Point graphql-to-mcp at a GraphQL endpoint and it auto-generates one MCP tool per query/mutation via introspection. Works with Claude Desktop, Cursor, Windsurf, and any MCP client.

Quick Start

Try it now — no install needed:

npx graphql-to-mcp https://countries.trevorblades.com/graphql

Or add to Claude Desktop / Cursor config:

{  "mcpServers": {    "countries": {      "command": "npx",      "args": ["-y", "graphql-to-mcp", "https://countries.trevorblades.com/graphql"]    }  }}

That's it. Claude can now query countries, continents, and languages.

Features

• Zero config — just provide a GraphQL endpoint URL
• Auto-introspection — discovers all queries and mutations automatically
• Flat parameter schemas — nested input objects are flattened for better LLM accuracy
• Smart truncation — large responses are intelligently pruned (array slicing + depth limiting)
• Auth support — Bearer tokens, API keys (header or query)
• Retry logic — automatic retries on 429/5xx with exponential backoff
• Include/exclude filters — expose only the operations you want
• Schema caching — skip re-introspection with --schema-cache for faster startup
• Mutation safety — auto-detect destructive mutations (delete*, remove*, etc.) and warn or block them

Usage

CLI

# Public API (no auth)npx graphql-to-mcp https://countries.trevorblades.com/graphql
# With bearer tokennpx graphql-to-mcp https://api.github.com/graphql --bearer ghp_xxxxx
# With API keynpx graphql-to-mcp https://api.example.com/graphql --api-key "X-API-Key:your-key:header"
# Filter operationsnpx graphql-to-mcp https://api.example.com/graphql --include "get*" --exclude "internal*"
# With prefix (avoid name collisions when using multiple APIs)npx graphql-to-mcp https://api.example.com/graphql --prefix myapi
# Cache schema locally for faster restartsnpx graphql-to-mcp https://api.example.com/graphql --schema-cache ./schema.json
# Force re-introspection (ignore cache)npx graphql-to-mcp https://api.example.com/graphql --schema-cache ./schema.json --force-refresh
# Block destructive mutations (delete*, remove*, etc.)npx graphql-to-mcp https://api.example.com/graphql --mutation-safety safe

Claude Desktop / Cursor Config

{  "mcpServers": {    "github": {      "command": "npx",      "args": [        "-y", "graphql-to-mcp",        "https://api.github.com/graphql",        "--bearer", "ghp_xxxxx",        "--prefix", "github"      ]    }  }}

Programmatic

import { createServer } from "graphql-to-mcp";
const server = await createServer({  endpoint: "https://api.example.com/graphql",  auth: { type: "bearer", token: "xxx" },  include: ["getUser", "listUsers"],});

How It Works

1. Introspect — Fetches the GraphQL schema via introspection query
2. Flatten — Nested InputObject types are flattened into simple key-value parameters (e.g., input.name → input_name)
3. Generate — Each query/mutation becomes an MCP tool with a flat JSON Schema
4. Execute — When an LLM calls a tool, the flat args are reconstructed into proper GraphQL variables and sent to your endpoint

Why Flat Schemas?

LLMs are significantly better at filling flat key-value parameters than deeply nested JSON objects. By flattening InputObject types, we get:

• Higher accuracy in parameter filling
• Fewer hallucinated nested structures
• Better compatibility across different LLM providers

Options

Smart Truncation

GraphQL APIs can return large payloads that overwhelm LLM context windows. graphql-to-mcp automatically:

• Slices arrays to 20 items (with metadata showing total count)
• Prunes depth beyond 5 levels (with object/array summaries)
• Hard truncates at 50K characters as a safety net

Schema Caching

Introspection queries can be slow on large schemas. Use --schema-cache to save the introspection result locally:

# First run: introspects and saves to cachenpx graphql-to-mcp https://api.example.com/graphql --schema-cache ./schema.json
# Subsequent runs: loads from cache (instant startup)npx graphql-to-mcp https://api.example.com/graphql --schema-cache ./schema.json
# Force re-introspection when the API schema changesnpx graphql-to-mcp https://api.example.com/graphql --schema-cache ./schema.json --force-refresh

The cache file stores the endpoint URL and timestamp. If you point at a different endpoint, it automatically re-introspects.

Mutation Safety

By default, graphql-to-mcp detects destructive mutations and adds warnings to their descriptions. This helps LLMs understand the risk before executing them.

Detected patterns: delete*, remove*, drop*, clear*, truncate*, destroy*, purge*, reset* (case-insensitive).

# Safe mode: only expose read queries + non-destructive mutationsnpx graphql-to-mcp https://api.example.com/graphql --mutation-safety safe
# Unrestricted: expose everything (use with caution)npx graphql-to-mcp https://api.example.com/graphql --mutation-safety unrestricted

Use with REST APIs Too

Pair with mcp-openapi to give Claude access to both REST and GraphQL APIs:

{  "mcpServers": {    "github-graphql": {      "command": "npx",      "args": ["-y", "graphql-to-mcp", "https://api.github.com/graphql", "--bearer", "ghp_xxx", "--prefix", "gh"]    },    "petstore-rest": {      "command": "npx",      "args": ["-y", "mcp-openapi", "https://petstore3.swagger.io/api/v3/openapi.json"]    }  }}

Related

• mcp-openapi — Same zero-config approach for REST/OpenAPI APIs

License

MIT