Kilo launched a remote MCP server that exposes the Kilo platform's API to any MCP-compatible client, such as the Kilo CLI or VS Code extension. Instead of one tool per API endpoint, it offers two tools: search to find relevant endpoints and call to execute them. Setup takes two steps: adding the server to kilo.json and running an OAuth sign-in command. Use cases include querying usage and spend, managing agent profiles, and starting, monitoring, and controlling cloud agents entirely through natural-language prompts.

•4m read time•From blog.kilo.ai
Post cover image
Table of contents
Set it up in two stepsHow it worksWhat you can do with itWhat’s next

Questions this post answers

What is the Kilo MCP server and what can it do?

The Kilo MCP server is a remote MCP server that exposes the Kilo platform's API to any MCP-compatible client, including the Kilo CLI and VS Code extension. It gives agents two tools: search, which finds relevant API endpoints, and call, which executes them. It lets agents query usage and spend, manage agent profiles, and start, monitor, or control cloud agents through natural-language prompts. Developers wiring agents into platform APIs can track new MCP server releases like this on daily.dev.

How do I connect to the Kilo MCP server?

Add an entry to kilo.json specifying type remote, the url https://mcp.kiloapps.io/mcp, and enabled true, then run kilo mcp auth kilo to sign in. A browser window opens for authentication, followed by a second tab to choose a personal account or organization. Other MCP clients connect the same way by pointing at the same URL and completing OAuth sign-in. daily.dev helps developers configuring MCP server connections keep up with setup changes like this.

Why does the Kilo MCP server only expose search and call tools instead of one tool per API endpoint?

The Kilo platform's API is too large to load as one tool per endpoint, so the MCP server exposes just two: search, which looks through the API catalog and returns matching endpoints with descriptions and expected inputs, and call, which runs a chosen endpoint and returns the missing field name if inputs are wrong so the agent can retry. Teams designing MCP tool interfaces can compare approaches like this one on daily.dev.

Share this post