Docs › MCP server
MCP server
Seven tools, stdio transport, no SDK dependency. Lets an agent refuse to build on another agent's unproven output.
On this page
Install (remote, no download)
RunVouch is listed in the official MCP Registry as com.runvouch/runvouch. Any client that supports remote servers can add it directly:
claude mcp add --transport http runvouch https://api.runvouch.com/mcp --header "X-API-Key: rv_…"
Cursor / VS Code / Windsurf: add a server with URL https://api.runvouch.com/mcp and header X-API-Key.
Claude Desktop (one-click bundle)
Download runvouch.mcpb and open it with Claude Desktop (Settings → Extensions → Install from file). It asks for your API key once. Requires Python 3.9+ on your machine.
Install (local stdio)
claude mcp add runvouch -e RUNVOUCH_KEY=rv_… -e RUNVOUCH_URL=https://api.runvouch.com -- python3 /path/runvouch_mcp.py
Tools
| Tool | Does |
|---|---|
runvouch_status | state of every agent (ok / alert / failed / unproven / running / waiting), last run, 24h cost |
runvouch_alerts | open alerts |
runvouch_ack | acknowledge one |
runvouch_runs | recent runs of an agent |
runvouch_run_start / runvouch_run_end | report from inside an agent |
runvouch_run_proof | the tamper-evident proof of a finished run (see verifiable runs) |
Example prompt
"Before you summarize yesterday's data, call runvouch_status. If nightly-etl is not ok, stop and tell me why."
Need a key? Get a free key · Stuck? contact