OpenAPI-MCP: Dockerized MCP Server to allow your AI agent to access any API with existing api docs

Generate MCP tool definitions directly from a Swagger/OpenAPI specification file.

OpenAPI-MCP is a dockerized MCP server that reads a swagger.json or openapi.yaml file and generates a corresponding Model Context Protocol (MCP) toolset. This allows MCP-compatible clients like Cursor to interact with APIs described by standard OpenAPI specifications. Now you can enable your AI agent to access any API by simply providing its OpenAPI/Swagger specification - no additional coding required.

Table of Contents

• Why OpenAPI-MCP?
• Features
• Installation
  • Using the Pre-built Docker Hub Image (Recommended)
  • Building Locally (Optional)
• Running the Weatherbit Example (Step-by-Step)
• Command-Line Options
  • Environment Variables

Demo

Run the demo yourself: Running the Weatherbit Example (Step-by-Step)

Why OpenAPI-MCP?

• Standard Compliance: Leverage your existing OpenAPI/Swagger documentation.
• Automatic Tool Generation: Create MCP tools without manual configuration for each endpoint.
• Flexible API Key Handling: Securely manage API key authentication for the proxied API without exposing keys to the MCP client.
• Local & Remote Specs: Works with local specification files or remote URLs.
• Dockerized Tool: Easily deploy and run as a containerized service with Docker.

Features

• OpenAPI v2 (Swagger) & v3 Support: Parses standard specification formats.
• Schema Generation: Creates MCP tool schemas from OpenAPI operation parameters and request/response definitions.
• Secure API Key Management:
  • Injects API keys into requests (header, query, path, cookie) based on command-line configuration.
    • Loads API keys directly from flags (--api-key), environment variables (--api-key-env), or .env files located alongside local specs.
    • Keeps API keys hidden from the end MCP client (e.g., the AI assistant).
• Server URL Detection: Uses server URLs from the spec as the base for tool interactions (can be overridden).
• Filtering: Options to include/exclude specific operations or tags (--include-tag, --exclude-tag, --include-op, --exclude-op).
• Request Header Injection: Pass custom headers (e.g., for additional auth, tracing) via the REQUEST_HEADERS environment variable.

Installation

Docker

The recommended way to run this tool is via Docker.

Using the Pre-built Docker Hub Image (Recommended)

Alternatively, you can use the pre-built image available on Docker Hub.

1. Pull the Image:

   docker pull ckanthony/openapi-mcp:latest

2. Run the Container: Follow the docker run examples above, but replace openapi-mcp:latest with ckanthony/openapi-mcp:latest.

Building Locally (Optional)

1. Build the Docker Image Locally:

   # Navigate to the repository root
   cd openapi-mcp
   # Build the Docker image (tag it as you like, e.g., openapi-mcp:latest)
   docker build -t openapi-mcp:latest .

2. Run the Container: You need to provide the OpenAPI specification and any necessary API key configuration when running the container.

   • Example 1: Using a local spec file and .env file:

     • Create a directory (e.g., ./my-api) containing your openapi.json or swagger.yaml.
     • If the API requires a key, create a .env file in the same directory (e.g., ./my-api/.env) with API_KEY=your_actual_key (replace API_KEY if your --api-key-env flag is different).

     docker run -p 8080:8080 --rm \\
         -v $(pwd)/my-api:/app/spec \\
         --env-file $(pwd)/my-api/.env \\
         openapi-mcp:latest \\
         --spec /app/spec/openapi.json \\
         --api-key-env API_KEY \\
         --api-key-name X-API-Key \\
         --api-key-loc header

     (Adjust --spec, --api-key-env, --api-key-name, --api-key-loc, and -p as needed.)

   • Example 2: Using a remote spec URL and direct environment variable:

     docker run -p 8080:8080 --rm \\
         -e SOME_API_KEY="your_actual_key" \\
         openapi-mcp:latest \\
         --spec https://petstore.swagger.io/v2/swagger.json \\
         --api-key-env SOME_API_KEY \\
         --api-key-name api_key \\
         --api-key-loc header

   • Key Docker Run Options:

     • -p <host_port>:8080: Map a port on your host to the container's default port 8080.
     • --rm: Automatically remove the container when it exits.
     • -v <host_path>:<container_path>: Mount a local directory containing your spec into the container. Use absolute paths or $(pwd)/.... Common container path: /app/spec.
     • --env-file <path_to_host_env_file>: Load environment variables from a local file (for API keys, etc.). Path is on the host.
     • -e <VAR_NAME>="<value>": Pass a single environment variable directly.
     • openapi-mcp:latest: The name of the image you built locally.
     • --spec ...: Required. Path to the spec file inside the container (e.g., /app/spec/openapi.json) or a public URL.
     • --port 8080: (Optional) Change the internal port the server listens on (must match the container port in -p).
     • --api-key-env, --api-key-name, --api-key-loc: Required if the target API needs an API key.
     • (See --help for all command-line options by running docker run --rm openapi-mcp:latest --help)

Running the Weatherbit Example (Step-by-Step)

This repository includes an example using the Weatherbit API. Here's how to run it using the public Docker image:

1. Find OpenAPI Specs (Optional Knowledge): Many public APIs have their OpenAPI/Swagger specifications available online. A great resource for discovering them is APIs.guru. The Weatherbit specification used in this example (weatherbitio-swagger.json) was sourced from there.

2. Get a Weatherbit API Key:

   • Go to Weatherbit.io and sign up for an account (they offer a free tier).
   • Find your API key in your Weatherbit account dashboard.

3. Clone this Repository: You need the example files from this repository.

   git clone https://github.com/ckanthony/openapi-mcp.git
   cd openapi-mcp

4. Prepare Environment File:

   • Navigate to the example directory: cd example/weather
   • Copy the example environment file: cp .env.example .env
   • Edit the new .env file and replace YOUR_WEATHERBIT_API_KEY_HERE with the actual API key you obtained from Weatherbit.

5. Run the Docker Container: From the openapi-mcp root directory (the one containing the example folder), run the following command:

   docker run -p 8080:8080 --rm \\
       -v $(pwd)/example/weather:/app/spec \\
       --env-file $(pwd)/example/weather/.env \\
       ckanthony/openapi-mcp:latest \\
       --spec /app/spec/weatherbitio-swagger.json \\
       --api-key-env API_KEY \\
       --api-key-name key \\
       --api-key-loc query

   • -v $(pwd)/example/weather:/app/spec: Mounts the local example/weather directory (containing the spec and .env file) to /app/spec inside the container.
   • --env-file $(pwd)/example/weather/.env: Tells Docker to load environment variables (specifically API_KEY) from your .env file.
   • ckanthony/openapi-mcp:latest: Uses the public Docker image.
   • --spec /app/spec/weatherbitio-swagger.json: Points to the spec file inside the container.
   • The --api-key-* flags configure how the tool should inject the API key (read from the API_KEY env var, named key, placed in the query string).

6. Access the MCP Server: The MCP server should now be running and accessible at http://localhost:8080 for compatible clients.

Using Docker Compose (Example):

A docker-compose.yml file is provided in the example/ directory to demonstrate running the Weatherbit API example using the locally built image.

1. Prepare Environment File: Copy example/weather/.env.example to example/weather/.env and add your actual Weatherbit API key:

   # example/weather/.env
   API_KEY=YOUR_ACTUAL_WEATHERBIT_KEY

2. Run with Docker Compose: Navigate to the example directory and run:

   cd example
   # This builds the image locally based on ../Dockerfile
   # It does NOT use the public Docker Hub image
   docker-compose up --build

   • --build: Forces Docker Compose to build the image using the Dockerfile in the project root before starting the service.
   • Compose will read example/docker-compose.yml, build the image, mount ./weather, read ./weather/.env, and start the openapi-mcp container with the specified command-line arguments.
   • The MCP server will be available at http://localhost:8080.

3. Stop the service: Press Ctrl+C in the terminal where Compose is running, or run docker-compose down from the example directory in another terminal.

Command-Line Options

The openapi-mcp command accepts the following flags:

Flag	Description	Type	Default
--spec	Required. Path or URL to the OpenAPI specification file.	string	(none)
--port	Port to run the MCP server on.	int	8080
--api-key	Direct API key value (use --api-key-env or .env file instead for security).	string	(none)
--api-key-env	Environment variable name containing the API key. If spec is local, also checks .env file in the spec's directory.	string	(none)
--api-key-name	Required if key used. Name of the API key parameter (header, query, path, or cookie name).	string	(none)
--api-key-loc	Required if key used. Location of API key: header, query, path, or cookie.	string	(none)
--include-tag	Tag to include (can be repeated). If include flags are used, only included items are exposed.	string slice	(none)
--exclude-tag	Tag to exclude (can be repeated). Exclusions apply after inclusions.	string slice	(none)
--include-op	Operation ID to include (can be repeated).	string slice	(none)
--exclude-op	Operation ID to exclude (can be repeated).	string slice	(none)
--base-url	Manually override the target API server base URL detected from the spec.	string	(none)
--name	Default name for the generated MCP toolset (used if spec has no title).	string	"OpenAPI-MCP Tools"
--desc	Default description for the generated MCP toolset (used if spec has no description).	string	"Tools generated from OpenAPI spec"

Note: You can get this list by running the tool with the --help flag (e.g., docker run --rm ckanthony/openapi-mcp:latest --help).

Environment Variables

• REQUEST_HEADERS: Set this environment variable to a JSON string (e.g., '{"X-Custom": "Value"}') to add custom headers to all outgoing requests to the target API.