Agent Tool Server
Learn how to deploy a custom tool server for AI agents with Ink MCP.
Create an Ink account and connect MCP
Make sure you have an Ink account and have connected Ink MCP to your agent. There are two ways to connect:
- OAuth — authenticate through a browser flow, no keys to manage
- Token — create an API key at ml.ink/account/api-keys
See Quick Start for full setup instructions.
Verify the connection
Once the MCP server is connected, confirm your agent can reach it by asking it to check.
Do you have Ink MCP connected?
Build the tool server
Ask your agent to build an HTTP API that other AI agents can use as a tool server. The agent will scaffold an Express app with endpoints for common agent operations.
Build an HTTP API that AI agents can use as a tool server. Include endpoints for web scraping, screenshot capture, and URL shortening. Use Node.js with Express.
Deploy with Ink
Tell your agent to deploy. Tool servers that handle scraping and screenshots benefit from more memory.
Deploy with Ink
create_service(
"name": "tool-server",
"repo": "ink/tool-server",
"build_pack": "railpack",
"port": "3000",
"memory": "512Mi"
)Use your tool server
Ink returns a live URL once the build completes. Point your AI agents to the deployed endpoints to use the tools.
Check logs and metrics (optional)
Ask your agent to pull build logs, runtime logs, or metrics for your deployed service.
Check metrics and logs
That's it — your agent tool server is deployed and live. From here you can add a custom domain, set up environment variables, or explore more examples.