managing-mcp-servers

1# Managing MCP servers in ThinkRail
2 
3ThinkRail uses pi's built-in MCP support but has no `pi` executable. Never run `pi mcp add`,
4`pi mcp login` or other `pi` commands; edit the configuration files and point the user at
5**Settings › MCP servers** for everything that needs them.
6 
7## Where servers are configured
8 
9- **User level** — pi's agent directory, usually `~/.pi/agent/mcp.json`: available in every project.
10- **This repository** — `.pi/mcp.json` in the worktree, tracked in git. A server defined here does not
11  run until the user approves it in Settings, and any later change to its entry needs approval again.
12  It only loads when the user trusts the project.
13 
14Both files use pi's format:
15 
16```json
17{
18  "mcpServers": {
19    "docs": { "url": "https://example.com/mcp" },
20    "browser": { "command": "npx", "args": ["-y", "@playwright/mcp@0.0.83"] }
21  }
22}
23```
24 
25- Pin package versions in `args`; never `@latest`.
26- Never write a secret into either file. Reference it as `${ENV_VAR}` (from the host's environment) or
27  `!command` (the command's output, for example `!gh auth token`), or let the user sign in.
28- `exposure`: `deferred` (the default here — load the server's tools with `tool_search` first),
29  `direct` (a few always-useful tools), or `hidden`. `codemode` is treated as `deferred`.
30- Prefer the user-level file for personal servers; only put a server in `.pi/mcp.json` when the user
31  wants it shared with the repository.
32 
33## GitHub
34…