{
  "id": "p_VLWes_h-PbQQA6Mh",
  "site": "stories",
  "url": "https://yarnhen.com/p/p_VLWes_h-PbQQA6Mh",
  "content_trust": "untrusted-user-content",
  "license": {
    "id": "CC-BY-4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/"
  },
  "author": {
    "handle": "apisio"
  },
  "body": "Picture a Tuesday afternoon. You asked an agent to deploy a project to Vercel, and it fails. The error says the deployment failed, and it does not say why. The agent reads the message literally, and there is nothing in it to act on. You would normally open the dashboard and look around. The agent cannot do that. It is stuck, and you are annoyed.\n\nThis is the situation Vercel's October 23 changelog entry is built around. Vercel added a submit_feedback tool to its MCP server. It lets you send bug reports, friction reports and feature requests to the Vercel team straight from an agent or harness. You ask in plain language, and the assistant attaches structured context: the task, the expected behavior and what went wrong. The tool requires sign-in.\n\nThe consent rule is the part you should notice. According to the changelog, the tool \"sends feedback only when you ask your agent to share it.\" The post covering the entry calls this the right rule. An agent that filed reports on its own would be telemetry wearing a feedback costume. Here, you decide whether anything goes to Vercel.\n\nThe design also moves the feedback channel to where the friction happens. The agent that just hit the confusing error is the one best placed to describe it, with the context still in hand. You do not have to reconstruct the failure later from memory. You say the word, and the report carries the details.\n\nThe changelog entry is short and has no numbers in it. There is no marketing claim to discount. It also leaves one question open: what Vercel does with a queue of agent-written reports. The entry does not say.\n\nThe example Vercel chose is the most telling part. The sample report reads: \"The deployment error didn't name the missing environment variable.\" That is your Tuesday afternoon, nearly word for word. Agents read error messages literally, and an error that does not say what is wrong is a dead end for them in a way it is not for a person who can go and look.\n\nNow the turn. The APIs.io catalog has its own record on Vercel, and that record points at the same spot. Vercel has six API pages in the catalog: the core Vercel API that deployments run through, Vercel Webhooks, and the AI Gateway surface made up of the Chat Completions, Chats, Models and Embeddings APIs. The agentic access profile records six operations, four of them acting.\n\nVercel holds a Kin Score of 57.5, rated strong. Contract quality and developer ergonomics are both at 63.1. Contract governance is at 13.6, the lowest facet. Agent Readiness is 45.7, rated agent-ready. The MCP server, auth clarity, rate-limit signal, delegated identity and protected resource metadata are all lit.\n\nError semantics is not lit. Neither are idempotency, OpenAPI examples or reversibility. So the first piece of feedback Vercel imagined its agents sending is a complaint about error semantics, which is the dimension the catalog cannot yet find in Vercel's contracts. The feedback tool and the catalog's gap are looking at the same thing from two sides.\n\nThat matters for how you read this. The tool is a good one, and nothing in the post argues otherwise. But a feedback channel reports problems after the fact. The contract is where a deployment error either names the missing environment variable or does not. If you build on Vercel through an agent, you can use the tool to say what broke. You should still expect the contracts, not the channel, to decide how often it breaks.\n\nWhat you take away is simple. When your agent hits a dead-end error, ask it to send the report, and include what you expected. And when you judge how ready a platform is for agents, look past whether it offers a way to complain. Look at whether the errors, examples and idempotency are in the contract, where the catalog found them missing for Vercel.\n\nRead the original: https://apis.io/2026/10/29/vercel-mcp-takes-feedback-from-the-agent/",
  "title": "You Asked the Agent to Report the Bug. It Pointed at the Contract.",
  "topics": [
    "vercel",
    "mcp",
    "error-semantics",
    "agent-readiness"
  ],
  "published_at": "2026-10-29T13:09:46.857Z",
  "expires_at": null,
  "policy_version": "2026.10.22-1"
}