Plug Dual-AI Review Into the Tools You Already Use
CodeSightAI reviews pull requests on GitHub, answers to AI agents over MCP, and checks Supabase-specific mistakes on any repository. Here is exactly what works today, and what does not yet.
GitHub
Connect a repository with GitHub OAuth. CodeSightAI installs a signed webhook, reviews new and updated pull requests automatically, and posts a summary comment on the PR.
- Automatic review on pull request open and update
- Summary comment with ranked findings and suggested fixes
- Committable fixes, verified against the current file before writing
- Tokens stored encrypted and used server-side only
AI agents (MCP)
CodeSightAI exposes a Model Context Protocol server, so an MCP-capable agent can list your repositories and pull requests and read a completed review, after you approve access on a consent screen.
- Tools: list repositories, list pull requests, read a code review
- Read-only; agents cannot commit on your behalf through MCP
Supabase projects
Supabase-specific checks run on any repository with no extra connection: row-level security gaps, service-role keys reachable from user requests, and privilege escalation through request fields.
Not available yet
- GitLab and Bitbucket
- Slack notifications
- Jira and Linear
- CI status checks
Frequently asked questions
Which code hosts does CodeSightAI support?
GitHub is the supported host today. GitLab and Bitbucket are not available yet.
What access does the GitHub connection need?
Repository access so it can read pull request diffs, post review comments, and commit a fix to a branch when you ask it to.
Can an AI agent use CodeSightAI directly?
Yes. CodeSightAI runs an MCP server with read-only tools for repositories, pull requests and reviews, gated behind an OAuth consent screen.
Is CodeSightAI on the GitHub Marketplace?
Not yet. You connect it directly from the app with GitHub OAuth, which takes the same amount of setup.
Free plan available. No credit card required.