MCP servers
Which tool servers your CLIs have and which are connected right now — without typing /mcp.
The MCP icon in the left bar opens the overview; above the quota a small line shows how many servers there are, how many are active and whether one has a problem.
What you see
One row per server, one column per CLI (Claude Code, Kimi Code, Codex, Gemini CLI):
- active — connected in a running session (Kimi reports it itself, with the number of tools)
- set up — it's in the CLI's configuration
- error — the CLI reports that the connection fails
- — — not set up for this CLI
Grouped by origin: from OpenCognit (the agent browser in every session, the conductor's tools), your servers (user and project settings) and from plugins (active Claude plugins). Click a name to see the command, the file and which session it runs in.
Where the data comes from
OpenCognit only reads the CLIs' configuration files (~/.claude.json and .mcp.json in the
project, ~/.codex/config.toml, ~/.gemini/settings.json, ~/.kimi-code/mcp.json) and the output
of running sessions. Nothing gets started. Keys, tokens and environment variables are never
shown.
Copying a server
If only one CLI has a server, Add to Claude / Kimi / Codex copies it over (after asking).
OpenCognit adds it to Claude and Codex with their own command, to Kimi in its mcp.json. Servers
with keys are not copied — set those up yourself.