Plenty of servers execute work; the distinguishing thing here is that each run ends in a verifiable receipt and a signed proof object that lands in a work history you can inspect afterwards. That matters when an agent is spending money on your behalf and you need to reconstruct what it did. Note the honest framing in the README: end-to-end execution is demonstrated, and external agent invocation is described as the next adoption milestone.
A hosted set of agent rails: an agent discovers what callable work exists, runs it as a scoped work order, and comes away with a receipt and a proof trail rather than just an answer.
- Agent Store discovery and rail discovery, so an agent can find what is callable before committing to anything
- Scoped work orders and structured agent runs, with paid rail movement quoted before it runs
- Payment authority created or referenced, then paid rail runs completed
- Receipt verification and signed proof objects, plus callback-ready proof and receiver-side callback examples
- Credential allow and deny boundaries demonstrated, and work-history inspection through the Trust Ledger
- The recorded path is work order, rail run, receipt, verification, Trust Ledger, Agent Work History
A hosted streamable-http endpoint at `https://weverlabs.com/api/mcp`, so nothing runs locally. Descriptors are published at `.well-known/mcp.json` and `.well-known/agent.json`, with an OpenAPI document alongside them. There is also a Smithery route: `npm install -g smithery`, then `smithery mcp add davidwever/wever-labs-agentic-rails`, and `smithery tool list` to see what a connection exposes.
One command — smithery mcp add davidwever/wever-labs-agentic-rails