SpecRun

Built with FastMCP for TypeScript.

✨ Features

• Zero Configuration: Filesystem is the interface - just drop OpenAPI specs in a folder
• Supports OpenAPI 3.0 and 2.0: Works with both OpenAPI 3.x and Swagger 2.0 specs
• Namespace Isolation: Multiple APIs coexist cleanly
• Full OpenAPI Support: Handles parameters, request bodies, authentication, and responses
• Run Any Tool to Interact with APIs: For example, cars_addCar to call POST /cars from cars.json spec to create a new car, or github_get_user_repos to call GET /user/repos from github.yaml spec to list repos.
• Run Any Tool with Custom Inputs: Pass structured JSON inputs for parameters and request bodies
• Run Any Tool to see Spec Details: Get the original OpenAPI spec details for any tool, including parameters, request body schema, and response schema
• Run Any Tool to get API responses as resources: Each tool call returns a JSON resource containing request URL, request body, and response
• Run Any Tool in Batch: One specrun_batch tool can execute any tool with multiple inputs and returns a consolidated JSON resource
• Auto Authentication: Simple .env file with {API_NAME}_API_KEY pattern
• Auto .env Placeholders: Adds {API_NAME}_SERVER_URL and {API_NAME}_BEARER_TOKEN entries when missing
• Multiple Transports: Support for stdio and HTTP streaming
• Built-in Debugging: List command to see loaded specs and tools
• MCP Prompts: Built-in prompts for listing tools, generating inputs, and explaining schemas
• Agent: configured agent for using SpecRun tools to explore and operate APIs in a guided way (.github/agents/specrun.agent.md)

Quick Start

Requirements

• Node.js 22 or newer

1️⃣ Install (optional)

npm install -g specrun

2️⃣ Create a specs folder where the server can read OpenAPI spec files. For example:

mkdir ~/specs

3️⃣ Add OpenAPI specs

Drop any .json, .yaml, or .yml OpenAPI specification files into your specs folder

4️⃣ Configure authentication (optional)

Create a .env file in your specs folder:

# ~/specs/.env
CARS_API_KEY=your_api_key_here

SpecRun will also ensure {API_NAME}_SERVER_URL and {API_NAME}_BEARER_TOKEN entries exist for each spec, adding empty placeholders when missing. When {API_NAME}_SERVER_URL has a value, SpecRun updates the spec file on load:

• OpenAPI 3.0: updates the first servers entry.
• OpenAPI 2.0 (formerly Swagger 2.0): updates host, schemes, and basePath (no servers section in OpenAPI 2.0).

SpecRun also watches the .env file and refreshes server URLs and auth config automatically after changes.

5️⃣ Add to MCP client configuration

Add to your MCP configuration:

If installed on your machine:

{
  "mcpServers": {
    "specrun": {
      "command": "specrun",
      "args": ["--specs", "/path/to/your/specs/folder"]
    }
  }
}

Otherwise:

{
  "mcpServers": {
    "specrun": {
      "command": "npx",
      "args": ["-y", "specrun", "--specs", "/absolute/path/to/your/specs"]
    }
  }
}

or with specific Node version:

{
  "mcpServers": {
    "specrun": {
      "command": "/Users/YOUR_USER_NAME/.local/bin/mcp-npx-node22",
      "args": ["specrun@latest", "--specs", "/absolute/path/to/your/specs"],
      "type": "stdio"
    }
  }
}

The mcp-npx-node22 script file uses nvm to run specrun with Node.js 22.14.0, ensuring compatibility regardless of the default Node version on your system.:

#!/bin/bash
# Set the PATH to include NVM's Node.js v22.14.0 installation
export PATH="/Users/YOUR_USER_NAME/.nvm/versions/node/v22.14.0/bin:$PATH"

# Execute npx with all passed arguments
exec npx "$@"

💻 CLI Usage

🚀 Start the server

# Default: stdio transport, current directory
specrun

# Custom specs folder
specrun --specs ~/specs

# HTTP transport mode
specrun --transport httpStream --port 8080

Run with Node 22 using npx

If your default node is older than 22, run SpecRun with Node 22 directly:

• npx -y node@22 ... runs the Node.js runtime, so the next argument must be a script path (for example ./node_modules/.bin/specrun).
• specrun@latest is an npm package spec and works directly with npx only when your current Node version already satisfies SpecRun requirements.

# Or list tools
npx -y node@22 ./node_modules/.bin/specrun list --specs ~/specs

# If your default Node is already 22+, this also works
npx -y specrun@latest --specs ~/specs

📋 List loaded specs and tools

# List all loaded specifications and their tools
specrun list

# List specs from custom folder
specrun list --specs ~/specs

🔑 Authentication Patterns

The server automatically detects authentication from environment variables using these patterns:

SpecRun also creates .env placeholders for:

If {API_NAME}_SERVER_URL is set, SpecRun writes that value into the spec before generating tools:

• OpenAPI 3.0: writes the first servers entry.
• OpenAPI 2.0 (formerly Swagger 2.0): writes host, schemes, and basePath.

Updates to .env are applied automatically without restarting the MCP server.

The {API_NAME} is derived from the filename of your OpenAPI spec:

• cars.json → CARS_API_KEY
• github-api.yaml → GITHUB_TOKEN
• my_custom_api.yml → MY_CUSTOM_API_KEY

🏷️ Tool Naming

Tools are automatically named using this pattern:

• With operationId: {operation_id}
• Without operationId: {method}_{path_segments}

Name normalization rules:

• Converted to snake_case
• Lowercased
• Non-alphanumeric characters normalized to _
• Truncated at the end when longer than 52 characters. (For VS Code/Copilot compatibility, stays within the practical 64-char internal limit.)
• Adds short suffixes only when needed to resolve collisions

Specs:

• get_car_by_id (from operationId)
• get_user_repos (generated from GET /user/repos)

Use the shared batch tool to run any tool with an array of inputs:

{
  "toolName": "cars_getCarById",
  "items": [{ "id": "123" }, { "id": "456" }],
  "failFast": false
}

Batch responses return a consolidated JSON resource with per-item outputs.

For batches over 200 items, SpecRun requires explicit confirmation. This is to prevent accidental large runs that could cause performance issues or unintended consequences. The server will return a message asking for confirmation, and you can retry with confirmLargeBatch: true and the provided confirmLargeBatchToken to proceed.

📦 Resource Outputs

Tool responses are returned as MCP resources with application/json content. Each resource includes:

1. Request URL
2. Request body
3. Response status and body

Example resource payload:

{
  "requestUrl": "https://api.example.com/v1/users/123",
  "requestBody": null,
  "response": {
    "status": 200,
    "body": {
      "id": "123",
      "name": "Jane Doe"
    }
  }
}

Batch runs return a single consolidated resource containing all item results.

📁 File Structure

your-project/
── specs/           # Your OpenAPI specs folder
   ├── .env            # Authentication credentials
   └── custom-api.yml  # Your OpenAPI spec files

🧭 MCP Prompts

SpecRun exposes MCP prompts for common workflows:

Detailed prompt guide with examples: PROMPTS_README.md

• list_apis: List loaded APIs/tools and ask the user to choose an endpoint
• generate_api_call: Generate a ready-to-run JSON input payload for a tool
• explain_api_schema: Explain parameters and request body schema with examples
• generate_random_data: Generate random ready-to-run JSON payload samples for a tool

📄 Example OpenAPI Spec

Here's a minimal example that creates two tools:

# ~/specs/example.yaml
openapi: 3.0.0
info:
  title: Example API
  version: 1.0.0
servers:
  - url: https://api-server.placeholder
paths:
  /users/{id}:
    get:
      operationId: getUser
      summary: Get user by ID
      parameters:
        - name: id
          in: path
          required: true
          schema:
            type: string
      responses:
        "200":
          description: User found
  /users:
    post:
      operationId: createUser
      summary: Create a new user
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              properties:
                name:
                  type: string
                email:
                  type: string
      responses:
        "201":
          description: User created