Integrations

Codna works where agents work.

One engine, everywhere your team already works: editor, terminal, CI and GitHub. The same codna commands run in every one of them.

DocsContact

Integrations

Cursor

Add Codna as an MCP server with codna mcp install --client cursor. Cursor's agent gets your repo mapped for zero tokens and fixes from an evidence bundle.

Claude

Expose five Codna tools to Claude over MCP: triage, fix, secure, recall and report. Recall runs on-device with no cloud round-trip.

GitHub App

Reviews every pull request with a clear verdict. Opens fix PRs from the codna-fix label, an @codna fix reply, or a red check suite. Never merges.

GitHub Action

thyn-ai/codna-action@v1 runs codna fix, review or secure in your CI. It opens pull requests and never merges. Pin package-spec to a codna version.

Model providers

Your agent. Your provider. Your key.

Codna maps your repo deterministically for zero tokens. The agent then runs on the model you choose, with your own key or the managed allowance.

CLIRun codna triage, fix, review and secure in any repo, CI job or container.
MCP serverGive Cursor and Claude five Codna tools: triage, fix, secure, recall and report.
GitHub App and ActionReview every pull request. Open fix PRs from an issue label, a comment or a red check.

Bring a key from Anthropic, OpenAI, Google Gemini, Groq, Mistral, OpenRouter or xAI. codna key set stores it in your OS keychain and never prints it.

Frequently asked

Install the mcp extra and run codna mcp install --client cursor or --client claude. Your editor gets five tools: codna_triage, codna_fix, codna_secure, codna_recall and codna_report_bug. The install helper never writes credentials.

Yes, when you ask. Label an issue codna-fix, reply @codna fix on a Codna finding with write access, or let a red check suite trigger a fix. The pull request states the issue, root cause, symbols and confidence. You review and merge.

Yes. thyn-ai/codna-action@v1 installs codna from PyPI and runs fix, review or secure with the same commands you run locally. Fix mode needs contents and pull-requests write; review mode needs contents read, pull-requests write and checks write.

No. Bring a key from Anthropic, OpenAI, Google Gemini, Groq, Mistral, OpenRouter or xAI, or sign in and use the managed allowance. With your own key you pay your provider directly.

Understanding runs on your machine and spends no tokens. Only the evidence bundle or the diff reaches your model provider. Set privacy.egress to fail-closed in codna.yaml and Codna runs your tests only under kernel-level network denial and makes no registry lookups during review.

One pip install. Python 3.10 or newer, macOS or Linux, git. No Node, Docker or server. Mapping a repo spends zero model tokens, and Codna prints the bundle size on every run.