api-mind

api-mind MCP Server

MCP server for API discovery from .mind spec files. Works with Claude Code, Claude Desktop, and any MCP-compatible AI assistant.

Quick Start

# 1. Install spec-mind to generate .mind files from OpenAPI specs
brew install spec-mind

# 2. Run the setup script from your project directory
curl -fsSL https://raw.githubusercontent.com/msegoviadev/api-mind-mcp/main/setup-mcp.sh | bash

# 3. Add your OpenAPI specs and generate .mind files
mkdir specs
cp your-api.yaml specs/
spec-mind sync --no-notation ./specs/

# 4. Restart Claude Code

In Claude Code: `` What APIs are available? Show me the endpoints for payments Call GET /payments/{id} in dev ``

---

Installation

Prerequisites

Install spec-mind to generate .mind files from your OpenAPI specs:

brew install spec-mind

Setup

Run setup-mcp.sh from your project directory:

curl -fsSL https://raw.githubusercontent.com/msegoviadev/api-mind-mcp/main/setup-mcp.sh | bash

Or for user-wide installation (all projects):

curl -fsSL https://raw.githubusercontent.com/msegoviadev/api-mind-mcp/main/setup-mcp.sh | bash -s -- --global

The script:

• Creates a specs/ directory in your project
• Registers api-mind in Claude Code with the correct specs path
• Scaffolds ~/.config/api-mind/dev.env for environment defaults

---

Usage

1. Add your API specs

# Copy your OpenAPI/Swagger YAML/JSON files to specs/
cp your-api.yaml specs/

2. Generate .mind files

spec-mind sync --no-notation ./specs/

3. Use in Claude Code

User: "What APIs are available?"
Claude: *uses list_apis tool*
"Found 2 APIs: ecommerce, payments"

User: "Show me payment endpoints"
Claude: *uses list_endpoints tool*
"POST /payments [auth: oauth2]
 GET /payments/{id}"

User: "Call GET /payments/{id} in dev"
Claude: *uses get_endpoint_schema + get_call_context tools*
"Resolved base URL: https://api.dev.example.com
 Calling GET /payments/123..."

---

Tools

list_apis

Lists all APIs loaded from the specs folder.

Input: none
Output: JSON with API names, titles, base URLs, and environments

list_endpoints

Lists endpoints across all APIs.

Input:
  filter (optional): Substring match on method, path, or section
Output: JSON with environments and endpoint list

get_endpoint_schema

Returns full context for an endpoint.

Input:
  api: API name
  method: HTTP method
  path: Endpoint path
Output: Text block with base URL, environments, auth, and schema

Call before constructing curl to understand the endpoint contract.

get_call_context

Returns runtime context needed to execute API calls.

Input:
  api: API name
  env (optional): Environment to use (dev, stage, uat). Defaults to dev.
Output: Resolved base URL, active environment, and default values for credentials and parameters

Call this before constructing curl when the user wants to actually invoke an endpoint. Reads from ~/.config/api-mind/<env>.env and ~/.config/api-mind/<api>/<env>.env.

---

Auth Patterns

When get_endpoint_schema shows auth requirements, construct headers:

| Auth in Schema | curl Header | |----------------|-------------| | None | No header | | bearer | -H 'Authorization: Bearer <TOKEN>' | | oauth2 <scopes> | -H 'Authorization: Bearer <TOKEN>' | | api_key <header> | -H '<header>: <KEY>' | | basic | -H 'Authorization: Basic <base64>' |

---

Environment Defaults

get_call_context reads default values from ~/.config/api-mind/:

~/.config/api-mind/
  dev.env        # base defaults for all APIs (dev environment)
  stage.env      # base defaults for stage
  auth0/
    dev.env      # API-specific overrides for auth0

Each .env file uses key=value format (lines starting with # are ignored). The base_url key overrides the placeholder URL from the spec.

Example ~/.config/api-mind/dev.env: `` base_url=https://api.dev.example.com auth0_client_id=abc123 auth0_cacert=/etc/ssl/cert.pem ``

setup-mcp.sh scaffolds this file on first run.

---

Workflow

list_apis → list_endpoints → get_endpoint_schema → get_call_context → [LLM constructs curl] → bash

1. list_apis - Discover available APIs
2. list_endpoints - Find relevant endpoints
3. get_endpoint_schema - Get endpoint contract (URL, auth, schema)
4. get_call_context - Resolve base URL and credentials for the target environment
5. LLM constructs curl command using resolved values
6. LLM executes via bash tool

---

Advanced Configuration

<details> <summary>Manual installation and team setup</summary>

Manual Installation

claude mcp add --transport stdio api-mind \
  --env SPECS_DIR=/absolute/path/to/specs \
  -- npx -y @msegoviadev/api-mind-mcp

Important: Always use absolute paths. MCP servers run as standalone processes without project context.

Project Configuration (.mcp.json)

Create .mcp.json in your project root:

{
  "mcpServers": {
    "api-mind": {
      "command": "npx",
      "args": ["-y", "@msegoviadev/api-mind-mcp"],
      "env": {
        "SPECS_DIR": "/absolute/path/to/specs"
      }
    }
  }
}

Note: Each team member needs their own .mcp.json with their absolute path. Add .mcp.json to .gitignore.

</details>

---

Development (Contributors)

git clone https://github.com/msegoviadev/api-mind-mcp
cd api-mind-mcp
npm install
npm run build

# Test locally
node dist/index.js /path/to/specs

---

Related

• spec-mind - Generate .mind files from OpenAPI specs
• api-mind - OpenCode plugin version

License

MIT