FeedLab + your AI agent

Let your agent do the testing legwork.

Connect Claude Code or any MCP client to FeedLab. It reads your codebase, writes test cases your team can run, and files the problems it finds as issues in your project.

Set it up in 5 minutes Browse the skills

What your agent can do

It always shows you its plan or findings first. Nothing is added to FeedLab until you agree.

Write test cases

Your agent reads the code for a feature and writes manual test cases your team can run: preconditions, steps, expected result, and priority.

Find and file issues

Ask for a code review and the agent reports real bugs, reliability, and performance problems as issues, each with a location and a suggested fix.

Run security reviews

A focused audit for broken access control, injection, secrets, and data exposure. Every finding comes with the attack that proves it.

Set up

You'll need a FeedLab account with a project, and an AI agent that supports MCP, such as Claude Code.

1. Create an API token
In FeedLab, go to Settings → API Tokens and create a token. It acts as you, so your agent can only reach projects you can. Copy it now; it's shown once.

2. Connect the FeedLab MCP server

In Claude Code, run this in your project folder with your token:

claude mcp add --transport http feedlab https://feedlab.cloud/api/mcp --header "Authorization: Bearer <your token>"

Using Cursor or another MCP client? Add this to its MCP config (for example .mcp.json):

{
  "mcpServers": {
    "feedlab": {
      "type": "http",
      "url": "https://feedlab.cloud/api/mcp",
      "headers": {
        "Authorization": "Bearer <your token>"
      }
    }
  }
}

3. Install the skills

Skills teach your agent to do each job well. This installs all 3 for Claude Code, once for every project. See what each one does.

mkdir -p ~/.claude/skills/feedlab-design-tests && curl -fsSL https://feedlab.cloud/skills/feedlab-design-tests/SKILL.md -o ~/.claude/skills/feedlab-design-tests/SKILL.md
mkdir -p ~/.claude/skills/feedlab-review-code && curl -fsSL https://feedlab.cloud/skills/feedlab-review-code/SKILL.md -o ~/.claude/skills/feedlab-review-code/SKILL.md
mkdir -p ~/.claude/skills/feedlab-security-review && curl -fsSL https://feedlab.cloud/skills/feedlab-security-review/SKILL.md -o ~/.claude/skills/feedlab-security-review/SKILL.md

4. Ask in plain language
Restart your agent so it picks up the server and skills, then ask.

Writing tests

From code to test cases

Name a feature, a branch, or the whole app. The agent maps every screen, action, rule, and role, drafts cases for the happy path, refusals, and edge cases, and compares them with the tests you already have. You approve the plan; the cases appear under Tests, ready to run in a session.

Try

Design tests for user authentication and add them to my project in FeedLab

After a change

Update the FeedLab tests for what I changed on this branch

Submitting issues

From review to triage

The agent reviews the code, checks each problem is real, and shows you the findings with severity and location. Approved findings land in the project's feedback, tagged code-review, next to reports from your users. Your team gets one notification per review, not one per issue.

When the agent fixes an issue, it can mark it resolved with a note linking the commit.

Code review

Review this codebase for bugs and reliability problems and register them as issues in my project in FeedLab

Security review

Run a security review of this codebase and register the vulnerabilities as issues in my project in FeedLab

Fix and close

Fix the high-severity FeedLab issues from the last review and mark them resolved

What the agent can access

The FeedLab MCP server gives agents these tools, and only for projects in your workspaces. There are no delete tools.

list_projectsSee the projects you have access to
list_testsRead a project's suites, groups, and test cases
create_test_suite · create_test_group · create_test_casesAdd tests, reusing existing suites and skipping duplicates
update_test_case · archive_test_caseKeep tests in step with the code
list_test_sessions · get_test_sessionRead session results to investigate failures
list_issuesSee open issues so nothing is filed twice
create_issuesRegister findings as issues, up to 25 at a time
update_issueMark an issue in progress or resolved once it's fixed

Staying in control

You approve first. The skills tell the agent to show its plan or findings and wait before writing anything.
Tokens act as you. Revoke one any time from Settings → API Tokens and it stops working immediately. Only a hashed copy is stored.
Your code stays with your agent. The agent reads code on your machine; FeedLab only receives the test cases and issues it creates.
Review what AI writes. Agents can be wrong. Treat tests and issues like a teammate's first draft.

How the agent's provider handles your code is covered by its own terms. See our Privacy Policy.

Ready to try it?

Create a free account, add a project, and connect your agent.

Get started free Explore the skills