autotask-mcp

A Model Context Protocol (MCP) server for Kaseya Autotask PSA, written in Go.

Provides AI assistants (Claude, GPT, etc.) with structured access to Autotask data and operations — tickets, companies, contacts, projects, time entries, billing, quotes, and more.

• 59 MCP tools covering all major Autotask entity types
• 4 lazy loading meta-tools for progressive tool discovery
• 7 MCP resource templates for direct data access
• Dual transport: stdio (Claude Desktop/Code) + HTTP (gateway deployments)
• Rate limiting: token bucket (5000 req/hour) + 3-thread concurrency limiter
• Circuit breaker: automatic backoff on API failures
• ID-to-name mapping cache with batch preloading and 30-minute per-entry TTL
• Compact response formatting to minimize LLM context usage

# Set credentials
export AUTOTASK_USERNAME=api_user@company.com
export AUTOTASK_SECRET=your_secret
export AUTOTASK_INTEGRATION_CODE=YOUR_CODE

# Run
go run .

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "autotask": {
      "command": "/path/to/autotask-mcp",
      "env": {
        "AUTOTASK_USERNAME": "api_user@company.com",
        "AUTOTASK_SECRET": "your_secret",
        "AUTOTASK_INTEGRATION_CODE": "YOUR_CODE"
      }
    }
  }
}

cp .env.example .env
# Edit .env with your credentials

# With Podman
podman-compose up -d

# With Docker
docker compose up -d

The server starts on port 8080 with endpoints:

• POST /mcp — MCP Streamable HTTP transport
• GET /health — Health check

For multi-tenant deployments, set AUTH_MODE=gateway. Credentials are injected per-request via headers:

X-API-Key: username
X-API-Secret: secret
X-Integration-Code: code

| Variable | Default