C++ coverage for AI agents

Help your AI agent improve test coverage.

Give your coding agent the same coverage you use in the editor. Let it find untested paths, write focused tests, and measure what changed.

Start free on public repositories or one private repository. Your agent can guide installation and sign-in.

  1. 01MeasureRun your real tests
  2. 02InspectFind uncovered paths
  3. 03ImproveAdd tests. Measure again.

Measure, inspect, and test

Find untested paths and add focused tests.

covdbg measures your existing Windows binary with its PDB. The MCP server gives the agent structured access to the results, without copying reports into a chat.

Find and run the tests

The agent reads covdbg's configuration guidance, discovers executables, and starts a coverage run. It can wait for completion or cancel the run.

guide → discover → run → wait_run

Read the gaps in context

Rank files by uncovered lines. Read the untested source segments with surrounding context, or ask a focused read-only SQL question.

files → code → query

Change, verify, repeat

Your agent edits tests using its coding tools. Run the tests and measure again to see whether the intended path was exercised. You review the change.

Edit tests → run → inspect

One command, your MCP client

Connect your coding agent with MCP.

The server runs locally over standard input and output as part of covdbg. Configure an MCP-capable coding agent to start it in your workspace.

covdbg mcp

Use your existing sign-in for development, or a project token for CI runs. The executable needs PDB symbols and a coverage configuration.

Read setup and tool reference
A prompt to start withAI agent

Read the covdbg guide, measure our test suite, and find an uncovered error path in our own code.

Add a focused test with meaningful assertions.

Run the tests again and report the coverage change, test result, and anything still untested.

The server measures and reads coverage. Your coding agent supplies the reasoning and test edits.

Review the tests as well as coverage.

A covered line does not prove a test has a useful assertion. A successful covdbg run also does not mean the tests passed. Keep the test result and the coverage result separate, and review what the agent changes.

Check how your AI client handles source code.

The server runs locally, but source context returned by the MCP tools is available to the connected AI client. That client's provider and settings determine where the context is processed.

Read the usage notes

Start with your own tests.

Free for public repositories and one private repository per person. Add team seats when you need more private repositories or private CI.

See plans and licensing

We store your theme and cookie choices to remember your preferences. With your permission, Google Analytics and our self-hosted Matomo help us understand how this website is used.

You can change your choice using Cookie settings in the footer. Cookie policy · Privacy policy