Built Riviera so you can fix how agents use your API without redeploying your backend

An agent picks the wrong tool, sends a valid request, and gets a successful response. Your logs look fine, but the user didn’t get what they asked for. Sometimes the problem is the tool description. “Creates an invoice” doesn’t explain when to use it, where to get the customer ID, or whether amounts are in dollars or cents. Fixing those words can mean a PR, review and deployment. So I built riviera.so: Import an OpenAPI document or connect an existing MCP server, then edit the tool and parameter descriptions agents read. Changes stay in a draft until you publish a version. Your backend still handles execution. Where this gets useful: > Making similar tools less ambiguous: Explain the difference between creating an invoice and recording a payment that already happened. > Fixing missing parameter context: tell the agent where an ID comes from, what units an amount uses, or which format a date expects. > Updating descriptions independently: review and publish wording changes without deploying your API or MCP server. >Investigating failures: see which tool was called, the published version, duration and outcome. >Turning mistakes into tests: Save a request and its expected tool, then compare tool selection against Production and your draft. One limitation worth calling out: model tests aren’t self-serve yet. I can help run them; they check tool selection without executing backend calls. If you’re running an MCP server, what’s harder today: getting agents to pick the right tool, or understanding what went wrong afterwards?