MCP server

Let Claude Code and other agents drive Tegment over MCP.

Tegment runs a local Model Context Protocol server that lets an MCP-aware agent (Claude Code, Gemini CLI, OpenCode, Antigravity CLI, and others) read your projects, worktrees, and issues, and start and tail your apps. It's how an agent working in a Tegment terminal can act on the rest of the app.

How it works#

Tegment runs a small local server on your machine while it's open. It listens on localhost only, with nothing exposed to the network, so your agents can reach it but nothing outside your machine can.

Agents connect through a helper that Tegment bundles, rather than talking to the server directly. If you run more than one Tegment window, an agent started in a terminal automatically reaches the window it belongs to.

Set it up#

Open Settings → AI Agents. Every agent has one card there, saying whether Tegment found it on this machine and whether it is set up, and a Setup button that registers the Tegment server for you. A card reading Partly set up tells you which half is missing and what you lose without it. (For the full per-agent picture, see Supported agents.)

• Claude Code: installing the Tegment plugin wires up the MCP server. The Setup button is the fallback if you don't use the plugin (it runs claude mcp add … tegment … mcp).
• OpenCode, Gemini CLI, Codex, Cursor, and Antigravity CLI: one click on Setup writes the Tegment server into the agent's own config for you.
• Other MCP clients: there's no auto-setup, so add it by hand. Point the client at the bundled tegment binary with the mcp argument; the Manual setup block at the bottom of the page shows the exact path.

Tools#

The server exposes a flat set of tools. There is no per-agent permission model, so a configured agent can use all of them.