AI-powered clustering & auto-PR generation

Your agents hit dead ends.
You never hear about it.

Add one file to your MCP server. Agents report what's missing. Patchwork generates pull requests that fix the gaps — automatically.

Get Started Free See How It Works

Free forever for small projects. No credit card required.

Open-source drop-in SDKs for every stack

Python TypeScript Go Rust

View on GitHub

Your MCP server is flying blind

When an agent can't find the right tool, it silently works around the problem — or gives up entirely. You have no idea what's failing, how often, or what to build next.

Missing tools go unnoticed

Agents improvise silently. You ship features nobody asked for while real gaps pile up.

No signal from the field

Agent logs are noisy and unstructured. There's no easy way to spot patterns across sessions.

Fixes take too long

Even when you know what's missing, going from insight to pull request is entirely manual.

your_mcp_server.py

from patchworkmcp import register_feedback

# One line. That's the whole integration.
register_feedback(mcp, api_key="pw_...")

# Your agents now have a submit_feedback tool.
# When they can't find what they need, they
# report it — automatically, with structure.

How it works

From setup to merged fix in four steps.

Add the feedback tool

Drop a single file into your MCP server. Agents gain a submit_feedback tool automatically.

Agents report gaps

When an agent can't find what it needs, it captures what happened, what was missing, and how it worked around it.

PRs are generated

Patchwork reads your repo, scores relevant files, and drafts a pull request that addresses the gap. One click.

Review & merge

You stay in control. Review the PR on GitHub, tweak if needed, merge, and your server is better.

Everything you need to close the loop

From raw feedback to merged PRs, Patchwork handles the full lifecycle.

Structured Feedback

Agents submit typed reports: what they needed, what gap type they hit, and how they worked around it.

AI Clustering

Similar feedback is automatically grouped into clusters. Heat scores rank issues by recency, frequency, and severity.

Auto PR Drafts

Select a feedback item and generate a pull request that addresses the gap. AI reads your source, scores relevant files, and drafts the fix.

Feedback Guards

Auto-dismiss noise, PII, and out-of-scope reports before they clutter your dashboard. System and per-server rules.

Agent Assist

When an agent submits feedback matching a known pattern, reply with guidance instantly. Improve agent experience without shipping code.

MCP-Native Management

Manage servers, review feedback, and create guards directly from your MCP client. OAuth-secured, fully audited.

"Can't I just read the agent logs?"

You could. Here's why teams don't.

Logs are noise

Thousands of lines, zero structure.

Agent logs mix successes and failures in a wall of text. Finding the moments where a tool was missing means reading everything.

Patchwork captures only the gaps, typed and structured for instant review.

No pattern detection

Each report looks like a one-off.

Without clustering, you can't tell the difference between a rare edge case and a systematic gap that 50 agents hit last week.

AI clustering groups similar feedback and ranks issues by heat score.

Fixing is manual

From "I see the problem" to PR: hours.

Even if you identify a gap in logs, you still have to find the right files, understand the codebase context, and write the fix yourself.

One click to generate a PR that reads your repo and drafts the fix.

Simple, transparent pricing

Start free. Scale when you're ready.

Unlimited team members on every plan. No per-seat fees, ever.

Monthly

Annual Save 17%

Free

$0 / forever

Perfect for solo devs exploring MCP feedback.

1 MCP server
500 feedback items / mo
Dashboard & analytics
5 LLM-generated improvements
2 PR drafts / mo
2 feedback guards
REST API access
30-day data retention

Get Started

Prefer to self-host?

Patchwork MCP is fully open source. Run the entire platform on your own infrastructure — same features, no limits, no vendor lock-in.

View on GitHub

Really?

Honestly, we’re built for indie developers and small teams. But if you genuinely need custom volumes, dedicated support, or on-premise deployment — let’s talk. No sales team, no demo theater.

Frequently asked questions

How do agents submit feedback?

Add a single file to your MCP server that registers a submit_feedback tool. When an agent can't find what it needs, it calls this tool with a structured report: what it needed, the gap type, and how it worked around the problem. SDKs are available for Python, TypeScript, Go, and Rust.

What's the difference between Dev and Published mode?

In Dev mode, every feedback item is shown individually and chronologically — perfect for early-stage servers where each report matters. In Published mode, feedback is automatically clustered by AI and ranked by heat score, so you see patterns instead of individual reports.

How does PR generation work?

Select a feedback item or cluster, click "Draft PR," and Patchwork reads your repository, scores files for relevance, and uses AI to generate a pull request that addresses the gap. The PR is created on GitHub for you to review and merge.

What are feedback guards?

Guards are rules that automatically filter feedback before it reaches your dashboard. System guards catch PII and credential leaks. Server guards let you dismiss known noise, duplicate reports, or out-of-scope requests. Every filtered item is still stored and auditable.

Can I manage everything from my MCP client?

Yes. Patchwork MCP exposes a full management interface via MCP tools. List servers, review trending issues, create guards, generate PRs — all from Claude Desktop, Cursor, or any MCP client. OAuth authentication keeps everything secure.

Is the free tier really free forever?

Yes. The free tier includes 1 server, 500 feedback items per month, the full dashboard, LLM-powered improvements, and PR drafts. No credit card required, no trial expiration. Upgrade when you need more servers, higher volume, or longer data retention.

Stop guessing what your agents need.

One file. Instant feedback. AI-powered fixes. Free forever for small projects.

Get Started Free

Your agents hit dead ends. You never hear about it.