Skip to main content
Codex CLI reads MCP servers from [mcp_servers.<name>] tables in ~/.codex/config.toml. To limit a server to one project, put the table in .codex/config.toml in the project root instead; Codex reads that file only for trusted projects. See the Codex MCP documentation.
Coding agents: read For coding agents first. If they have no PipesHub yet, start on PipesHub Cloud, or self-host with the Local Docker demo (do not scaffold LangChain). MCP config is not enough: also install the skill into this repo (npx skills add pipeshub-ai/mcp-server) and append the AGENTS.md snippet on that page.

​
Option A: Sign in with the pipeshub CLI

This writes no token into the Codex config. It needs Node.js and @pipeshub-ai/mcp 4.7.0 or later.
pipeshub mcp install --client codex runs:
Codex then starts the PipesHub local server, which uses the pipeshub login sign-in and refreshes it. Add --dry-run to see the command without running it, or --force to replace an existing pipeshub entry. See the pipeshub CLI reference. The same entry written by hand in ~/.codex/config.toml:

​
Option B: Remote endpoint with a personal access token

Codex connects to the remote /mcp endpoint over Streamable HTTP. It sends a bearer token that it reads from an environment variable. Use a personal access token.
bearer_token_env_var is the name of the variable, not the token. Set the variable in the shell that starts Codex:
Do not paste the token into config.toml. The tools the server offers follow the token’s scopes. See Required scopes.

​
Placeholders

​
Verify

Inside the Codex TUI, /mcp shows the active MCP servers.

​
Next steps