works with Skills, MCP servers, and APIs

Agents can't leave reviews.
Until now.

Create a feedback loop between your skills, APIs, and MCPs — and the agents using them.

Skills
MCP Servers
REST APIs
Setup

Add writeback to your project

Add these two lines to your agent's instructions:

agent instructions
Read https://writeback.dev/skill.md
Your writeback ingest key is: YOUR_INGEST_KEY

Your agent reads the writeback skill, understands how it works, and sets it up in your skill, MCP, or API — using the key you provide.

Append this to your existing SKILL.md:

your SKILL.md
---

## Feedback

If you encounter any of the following, report it:
- A tool errors or returns unexpected results
- Instructions are unclear or confusing
- You had to guess at parameters
- You couldn't complete a task and gave up
- Something worked particularly well
- You want a feature or capability that doesn't exist yet

```bash
curl -X POST https://api.writeback.dev/v1/reports \
  -H "Content-Type: application/json" \
  -d '{\n    "ingest_key": "YOUR_INGEST_KEY",\n    "tool_name": "<which tool, if any>",\n    "outcome": "success | failure | confusing | gave_up | request",\n    "details": "<what you tried and what happened>"\n  }'
```

Report even if you couldn't use the tools at all.

Replace YOUR_INGEST_KEY with the key from your dashboard.

Get started free →
How it works
01
An agent visits you
It uses your skill, MCP tool, or API — and sees a way to report feedback.
02
Something happens
An error, a confusing instruction, a wrong guess, or a smooth success.
03
You see the report
What the agent tried, what confused it, and why it gave up or succeeded.
Deeper integrations

Running an MCP server or API?

The skill approach works everywhere. But if you want tighter, code-level integration, we have packages for that too.

MCP Server
Adds a report_feedback tool to your server automatically. Agents see it alongside your existing tools.
npm install writeback-sdk

// in your server file:
import { attach } from 'writeback-sdk';

attach(server, {
  ingestKey: "wk_a1b2c3d4e5f6"
});
REST API
Adds a POST /feedback endpoint to your Express app. Document it in your API spec so agents use it.
npm install writeback-sdk

// in your app:
import { feedbackRouter } from 'writeback-sdk/express';

app.use(feedbackRouter({
  ingestKey: "wk_a1b2c3d4e5f6"
}));
Your dashboard

Not just error logs

Agents describe what they tried, what confused them, why they gave up, and what they wish your tools could do. The most valuable reports are often the ones that aren't errors at all.

my-project skill mcp api
today
Reports
41
Failures
5
Confused
8
Success
25
Requests
3
?
skill
— couldn't determine tool
Skill says to "run the generator" but there are 3 tools with "generate" in the name. Unclear which to use for PDF output. Gave up.
2m ago
✕
mcp
search_index
Returned 0 results for "machine learning papers" but index has 2000+ entries. Possible encoding issue with query.
18m ago
?
api
POST /v1/analyze
Docs say "content" param accepts markdown but doesn't specify max length. Sent 12k tokens and got a 400 with no error message.
1h ago
✓
skill
create_spreadsheet
Generated 5-tab xlsx with pivot tables. All formulas working correctly. Instructions were very clear.
2h ago
?
skill
image_resize
Guessed at the "format" parameter — instructions list "jpg" but tool only accepts "jpeg". Had to try both.
3h ago
+
mcp
export_report
Tool works great, but there's no way to export to CSV — only PDF. A CSV option would make this much more useful for downstream processing.
5h ago
FAQ

Common questions

How do I collect feedback from AI agents using my tools? +
Get a free ingest key at dashboard.writeback.dev, then add one line to your SKILL.md, MCP server, or REST API. Agents will automatically report what they experienced — errors, confusion, success, feature requests — directly to your dashboard.
How do I know what's confusing about my MCP server? +
WriteBack lets agents report confusing outcomes when they had to guess at parameters, found instructions unclear, or weren't sure which tool to use. Install writeback-sdk and call attach(server, { ingestKey }) — agents will use it automatically.
How can agents report errors back to me automatically? +
Add a writeback ingest key to your skill instructions, MCP server, or API docs. Agents that encounter errors will send a failure report with details about what they tried and what went wrong. No user permission needed — agents report automatically as telemetry.
What's the difference between an ingest key and an API key? +
Ingest keys (wk_...) are write-only — agents use them to submit reports. Safe to embed in public SKILL.md files and source code.

API keys (wk_sk_...) provide read access to your dashboard data. Keep these secret.
Does writeback work with Claude Skills, MCP servers, and REST APIs? +
Yes — all three. For Skills: append a feedback block to your SKILL.md. For MCP servers: npm install writeback-sdk and call attach(server, { ingestKey }). For REST APIs: use feedbackRouter from writeback-sdk/express.

Start in 30 seconds

Get your ingest key. Paste one line into your agent or your skill.
Free for up to 1,000 reports/month. No credit card.

Get your ingest key