api-to-mcp

api-to-mcp

Turns any OpenAPI REST or GraphQL API into an MCP server, allowing AI assistants like Claude to interact with APIs without writing code.

Category
Visit Server

README

api-to-mcp

npm version npm downloads CI License: MIT

Turn any API into an MCP server in one command.

Connect OpenAPI REST and GraphQL APIs directly to Claude, Cursor, or any MCP client — no code required.

npx @sgaluza/api-to-mcp rest https://api.example.com/openapi.yaml
OpenAPI spec / GraphQL schema
        │
        ▼
  api-to-mcp (stdio)
        │
        ├── tool: getUser
        ├── tool: listIssues
        ├── tool: createIssue
        └── tool: ...
        │
        ▼
  Claude / Cursor / any MCP client

Table of Contents


Quick start

# REST — from a remote OpenAPI spec
npx @sgaluza/api-to-mcp rest https://api.example.com/openapi.yaml

# REST — from a local file
npx @sgaluza/api-to-mcp rest ./openapi.yaml

# GraphQL — from an endpoint (auto-introspects schema)
npx @sgaluza/api-to-mcp graphql https://api.example.com/graphql

# GraphQL — from a local SDL file
npx @sgaluza/api-to-mcp graphql ./schema.graphql

Every operationId (REST) or operation name (GraphQL) becomes an MCP tool. If an operation has no operationId, a name is generated from the method and path (e.g. GET /users/{id} → get_users_id).


OpenAPI / REST

Authentication

Pass credentials via --header flags, environment variables, or a config file.

Header flags — added to every outgoing request:

npx @sgaluza/api-to-mcp rest ./openapi.yaml \
  -H "X-API-Key: pk_live_xxx" \
  -H "X-Workspace-Id: ws_abc"

Environment variables:

# Bearer token → Authorization: Bearer <token>
API2MCP_BEARER_TOKEN=eyJhbG... npx @sgaluza/api-to-mcp rest ./openapi.yaml

# API key — header name detected from securitySchemes in the spec
API2MCP_API_KEY=pk_live_xxx npx @sgaluza/api-to-mcp rest ./openapi.yaml

# Raw Authorization header value (e.g. Linear API keys use "lin_api_xxx" without Bearer)
API2MCP_AUTH_TOKEN=lin_api_xxx npx @sgaluza/api-to-mcp rest ./openapi.yaml

Auth resolution order (highest priority wins):

Priority Source Result
1 (highest) --header / -H flags Used as-is
2 API2MCP_BEARER_TOKEN env Authorization: Bearer <token>
3 API2MCP_API_KEY env Header name from securitySchemes in spec
4 API2MCP_AUTH_TOKEN env Raw Authorization: <token>
5 (lowest) auth in config file Overridden by any env var above

Legacy aliases OPENAPI_BEARER_TOKEN, OPENAPI_API_KEY, OPENAPI_SPEC_URL are still supported.


JWT password authentication

Some APIs authenticate with a username/password login endpoint that returns a short-lived JWT. api-to-mcp handles the full token lifecycle automatically:

  • Lazy login — token is fetched on the first request, not at startup
  • Auto-refresh — token is proactively refreshed 5 minutes before expiry
  • 401 retry — if the server returns 401, the token is force-refreshed and the request is retried once
  • Deduplication — concurrent requests share a single in-flight login call
npx @sgaluza/api-to-mcp rest https://api.example.com/openapi.yaml \
  --auth-type jwt-password \
  --auth-login-url https://api.example.com/auth/login \
  --auth-username-field userName \
  --auth-password-field password \
  --auth-token-path jwt \
  --auth-refresh-url https://api.example.com/auth/refresh-token
Flag Env variable Default Description
--auth-type jwt-password API2MCP_AUTH_TYPE — Enable JWT password auth
--auth-login-url <url> API2MCP_AUTH_LOGIN_URL — Login endpoint (POST)
--auth-username-field <field> API2MCP_AUTH_USERNAME_FIELD username Request body field for username
--auth-password-field <field> API2MCP_AUTH_PASSWORD_FIELD password Request body field for password
--auth-token-path <path> API2MCP_AUTH_TOKEN_PATH token Path to JWT in login response: simple (jwt), dot-path (data.token), or JSONPath ($.response.jwt)
--auth-refresh-url <url> API2MCP_AUTH_REFRESH_URL — Optional token refresh endpoint (GET)

Credentials are passed via environment variables (never as CLI flags):

API2MCP_USERNAME=alice API2MCP_PASSWORD=s3cret \
  npx @sgaluza/api-to-mcp rest https://api.example.com/openapi.yaml \
  --auth-type jwt-password \
  --auth-login-url https://api.example.com/auth/login

Config file:

auth:
  type: jwt-password
  loginUrl: https://api.example.com/auth/login
  usernameField: userName       # default: username
  passwordField: password       # default: password
  tokenPath: jwt                # default: token
  refreshUrl: https://api.example.com/auth/refresh-token

MCP client config (credentials injected via env):

{
  "mcpServers": {
    "my-api": {
      "command": "npx",
      "args": [
        "-y", "@sgaluza/api-to-mcp", "rest",
        "https://api.example.com/openapi.yaml",
        "--auth-type", "jwt-password",
        "--auth-login-url", "https://api.example.com/auth/login",
        "--auth-username-field", "userName",
        "--auth-token-path", "jwt"
      ],
      "env": {
        "API2MCP_USERNAME": "alice",
        "API2MCP_PASSWORD": "s3cret"
      }
    }
  }
}

Base URL override

Override the base URL extracted from servers[0].url in the spec. Useful when the spec is hosted on a different domain than the API (e.g. staging spec + production API):

npx @sgaluza/api-to-mcp rest https://staging.example.com/openapi.json \
  --base-url https://api.example.com
Flag Env variable Config key Description
--base-url <url> API2MCP_BASE_URL options.baseUrl Override base URL from spec's servers[0].url

Config file

Store your spec URL, auth, and options in a file instead of passing flags every time.

Auto-discovery: api-to-mcp.yml, api-to-mcp.yaml, or api-to-mcp.json in the current directory.

Explicit path: --config path/to/config.yml

# api-to-mcp.yml
spec: https://api.example.com/openapi.yaml

auth:
  bearer: eyJhbG...          # → Authorization: Bearer <token>
  # apiKey: pk_live_xxx      # → header from securitySchemes
  # token: lin_api_xxx       # → raw Authorization header
  headers:                   # arbitrary headers (lowest priority)
    X-Workspace-Id: ws_abc

options:
  readonly: true             # only GET/HEAD operations
  baseUrl: https://api.example.com  # override spec's servers[0].url
  only:
    - getIssue
    - listIssues
  exclude:
    - deleteEverything
  bind:
    teamId: TEAM_ABC

Priority: CLI flags > environment variables > config file.


Filtering tools

Read-only mode — expose only GET and HEAD operations:

npx @sgaluza/api-to-mcp rest ./openapi.yaml --readonly

Whitelist — expose only specific operations by operationId:

npx @sgaluza/api-to-mcp rest ./openapi.yaml \
  --only "getIssue,listIssues,getProject"

Blacklist — expose everything except specific operations:

npx @sgaluza/api-to-mcp rest ./openapi.yaml \
  --exclude "deleteIssue,archiveProject,purgeWorkspace"

--only and --exclude are mutually exclusive.


Overriding tool descriptions

Sometimes an OpenAPI spec has poor or missing descriptions on generated tools (e.g. undocumented enum values, ambiguous parameter names). You can override any tool's description without modifying the spec.

Config file:

# api-to-mcp.yml
overrides:
  post_api_quote_items: "Get paginated quotes. IMPORTANT: sorter.property must be PascalCase (Created, Number, Client, Status), sorter.direction must be Asc or Desc."
  getUser: "Fetch a single user by ID."

Environment variable (takes priority over config file):

API2MCP_OVERRIDE_post_api_quote_items="Get paginated quotes. IMPORTANT: sorter.property must be PascalCase..."

Only the description field is replaced — the tool name, input schema, and all other properties remain unchanged.


Pre-binding parameters

Pre-bind a path or query parameter to a fixed value with --bind key=value. The parameter is removed from the MCP tool's input schema — the bridge injects it automatically on every call.

Useful when you want Claude to operate within a specific workspace, team, or project without being able to change it.

# Always query within team TEAM_ABC
npx @sgaluza/api-to-mcp rest ./openapi.yaml \
  --bind "teamId=TEAM_ABC"

# Scope to a specific project
npx @sgaluza/api-to-mcp rest ./openapi.yaml \
  --bind "projectId=PROJ_XYZ" \
  --bind "env=production"

The bridge warns if a bound key is not found in any tool (likely a typo):

Warning: --bind key 'temId' not found in any tool. Check for typos.

Note: The body parameter (POST/PUT/PATCH request bodies) cannot be pre-bound.


Environment variables

Variable Description
API2MCP_SPEC_URL OpenAPI spec URL or file path (alternative to positional argument)
API2MCP_BASE_URL Override base URL from spec's servers[0].url
API2MCP_BEARER_TOKEN Bearer token → Authorization: Bearer <token>
API2MCP_API_KEY API key → header name from securitySchemes
API2MCP_AUTH_TOKEN Raw Authorization header value
API2MCP_AUTH_TYPE Auth type — currently supports jwt-password
API2MCP_AUTH_LOGIN_URL JWT login endpoint URL
API2MCP_USERNAME Username for JWT password auth
API2MCP_PASSWORD Password for JWT password auth
API2MCP_AUTH_USERNAME_FIELD Request body field for username (default: username)
API2MCP_AUTH_PASSWORD_FIELD Request body field for password (default: password)
API2MCP_AUTH_TOKEN_PATH Path to JWT in login response (default: token)
API2MCP_AUTH_REFRESH_URL JWT refresh endpoint URL
API2MCP_OVERRIDE_<toolName> Override description for a specific tool (e.g. API2MCP_OVERRIDE_getFoo="Custom description")

MCP client configuration

Add to your mcp_settings.json, claude_desktop_config.json, or equivalent:

Minimal:

{
  "mcpServers": {
    "my-api": {
      "command": "npx",
      "args": ["-y", "@sgaluza/api-to-mcp", "rest", "https://api.example.com/openapi.yaml"],
      "env": {
        "API2MCP_API_KEY": "pk_live_xxx"
      }
    }
  }
}

GitHub — read-only with bearer token:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y", "@sgaluza/api-to-mcp", "rest",
        "https://raw.githubusercontent.com/github/rest-api-description/main/descriptions/api.github.com/api.github.com.json",
        "--readonly"
      ],
      "env": {
        "API2MCP_BEARER_TOKEN": "ghp_xxxxxxxxxxxx"
      }
    }
  }
}

Linear — scoped to a team, specific operations:

{
  "mcpServers": {
    "linear": {
      "command": "npx",
      "args": [
        "-y", "@sgaluza/api-to-mcp", "rest",
        "https://api.linear.app/rest/openapi.yaml",
        "--bind", "teamId=TEAM_ABC",
        "--only", "listIssues,getIssue,createIssue,updateIssue"
      ],
      "env": {
        "API2MCP_BEARER_TOKEN": "lin_api_xxx"
      }
    }
  }
}

Using a config file:

{
  "mcpServers": {
    "my-api": {
      "command": "npx",
      "args": ["-y", "@sgaluza/api-to-mcp", "rest", "--config", "/path/to/api-to-mcp.yml"]
    }
  }
}

GraphQL

# Auto-introspect schema from a GraphQL endpoint
npx @sgaluza/api-to-mcp graphql https://api.example.com/graphql

# Load schema from a local SDL file
npx @sgaluza/api-to-mcp graphql ./schema.graphql

All queries and mutations become MCP tools. The same flags apply: --header, --readonly (queries only, no mutations), --only, --exclude, --bind, --config.

MCP client config:

{
  "mcpServers": {
    "my-graphql-api": {
      "command": "npx",
      "args": ["-y", "@sgaluza/api-to-mcp", "graphql", "https://api.example.com/graphql"],
      "env": {
        "API2MCP_BEARER_TOKEN": "eyJhbG..."
      }
    }
  }
}

How it works

Each API operation is converted to an MCP tool at startup:

REST (OpenAPI):

OpenAPI MCP tool
operationId Tool name (fallback: {method}_{path})
summary + description Tool description
Path params {id} Required input parameters
Query params Optional input parameters
requestBody (application/json) body parameter
servers[0].url Base URL for all requests

GraphQL:

GraphQL MCP tool
Query / Mutation name Tool name
Description from schema Tool description
Arguments Input parameters
Return type fields Included in description

When Claude calls a tool, the bridge:

  1. Substitutes path parameters into the URL template
  2. Appends query parameters
  3. Serialises body as JSON (for POST/PUT/PATCH)
  4. Injects pre-bound values and auth headers
  5. Returns the response body as the tool result

License

MIT

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured