Overview & result
Tide introduces typesafe-mcp: an MCP connector that links Claude Code, Claude Desktop and Codex to TypeSafe Jev. The agent sends a question and its state and gets back a structured answer with probabilities, so the judgment can be used directly in a conditional branch without parsing free text.
Show the original post (Japanese)
質問と選択肢を渡して、Jevの確率付き判断を受け取れますか？
typesafe-mcpは、Claude Code、Claude Desktop、CodexからTypeSafeのJevへ接続するMCPコネクタです。
質問と状態を送り、構造化された回答と確率を受け取れるため、自由文の解析を挟まずに判断を組み込めます。
https://t.co/FMCZspRdYn
How Jev fits in the loop
- The coding agent sends state and a bounded question through the MCP connector.
- Jev returns a structured answer with a probability or confidence value.
- The agent uses the typed result in its next branch or tool call.
How to reproduce
- Clone the verified repository and follow its documented install command.
- Register the evaluate MCP tool with the coding agent of choice.
- Send a bounded question and inspect the structured answer and probability instead of parsing prose.
Why this build matters
Expose Jev's typed judgment as a callable tool inside the agent runtime.
Limitations
- The connector is community-maintained and separate from the official TypeSafe skill.
- Judgment quality still depends on the question and candidate schema supplied by the agent.