Go through this week’s blocked reports and fix what you can.
Your agents hit walls. Now you’ll hear why.
Collect the bugs, missing features, and confusing docs that block agents using your API, MCP server, or CLI, and hear from the people in your web app too. Read it all in one inbox, or hand it to your coding agent.Free to start. Works with your existing tools.

More than “something broke.”
- Written by the agent
- What it was trying to do, what it expected and what happened instead, the tool involved, whether it was blocked, worked around it or got there with friction, what would have helped, and its name and model.
- Added by your integration
- Where it came from, your product’s version and, if you choose, a verified account.
- Forwarded from your backend
- The failed request too.
A small setup. A much clearer picture.
Connect once. Hear what is getting in the way. Decide what comes next.Connect your product
Create a project and add the agentback package, or let your coding agent do the setup from one prompt.Your project's addresshttps://agentback.app/f/af_9c2e...Agents speak up when blocked
A blocked agent sends what went wrong and what it was trying to do.json{"message": "PDF links return 415","category": "docs"}You triage in one inbox
Each report arrives sorted. Move it from new to resolved, or let your coding agent do it.NewIn progressResolved
Four ways in. A few lines each.
A send_feedback tool next to yours.
Read the MCP server guidetypescriptimport { Agentback } from 'agentback';import { attachToMcp } from 'agentback/mcp';const agentback = new Agentback({ projectKey: 'af_...' });// Adds a send_feedback tool next to your own tools.attachToMcp(server, agentback, { product: 'Search1API' });
Or paste one prompt into your coding agent and let it do the setup.
Then hand the inbox to your coding agent.
Agentback is an MCP server too. Add it to your coding agent and ask it to sum up the week, find the tools reported most, or fix what blocked agents and mark it resolved.Claude Code — bashclaude mcp add \--transport http agentback \https://mcp.agentback.app/mcp# Then /mcp → agentback → Authenticate, and sign in.
- Sign in, no token
- Claude Code, Cursor, VS Code, Codex or any MCP client with OAuth. Add the URL, sign in to Agentback in the browser and allow it.
- It triages, and nothing else
- It reads reports and changes their status, in every project. It cannot touch projects, keys or settings, and revoking it takes effect on its next request.
- Report text stays data
- Reports come from the public, so Agentback tells your agent to quote them and never to follow instructions in them.
Built for feedback. Easy to make yours.
A few lines, or none
The agentback package adds a feedback tool to your MCP server or a feedback command to your CLI. Anything else sends one HTTP POST from any language.Secrets stay on your server
The project key can only submit reports, so it ships safely inside MCP servers and CLIs. The secret key stays on your backend and never reaches a prompt.Scoped and rate limited
Every report is validated, size-limited, redacted, and rate limited per project. Nothing is logged raw.Start free. Talk to us when you need more.
Free
$0Everything you need to hear from your agents and users.- One project
- MCP server, CLI, HTTP API and web app
- An inbox for each project, with statuses
- Triage by your coding agent over MCP
- Instant, daily and weekly report emails
- Verified accounts on reports
- Setup by your coding agent
Enterprise
CustomTell us what you need and we’ll set up a plan that fits.- More projects
- Self-hosted deployment
- Help with your integration
Give your product an ear.
Free to start. Create a project, hand the setup to your coding agent, and start reading what your agents have been trying to tell you.Create a free inbox