GitHub · by Mostafa

typesafe-mcp

An MCP connector that gives your agent direct access to Jev.

Mostafa’s MCP connector puts Jev inside Claude Code, Claude Desktop and Codex. The agent can call Jev for Choice, Score and Noul judgments whenever it needs one.

The problem it names is specific. When an agent needs a yes-or-no call, a routing decision or a severity rating, it usually asks a language model, parses the prose and hopes the format holds. The answer arrives with no probability attached, so the agent cannot tell a confident yes from a coin flip. This server exposes a single `evaluate` tool that sends state plus typed questions and returns structured answers with probabilities. There is nothing to parse and no prompt formatting to maintain.

One command, `evaluate setup mcp`, registers it with Claude Desktop, Claude Code and Codex where they are detected, and `evaluate setup pi` covers pi. It is a single Go binary that installs to `~/.local/bin` and updates itself in place. Requests retry on 429 and 529.

Either key works: with `TYPESAFE_API_KEY` it calls TypeSafe’s own endpoint, and with `OPENROUTER_API_KEY` it routes through OpenRouter’s Decisions endpoint and bills that account instead. `TYPESAFE_BASE_URL` points the TypeSafe route at a proxy or a self-hosted gateway. Mostafa notes that OpenRouter’s Decisions endpoint is still on an alpha path and may move.