Skip to main content
The CodeRabbit MCP server builds most of its tools from the CodeRabbit API. A tool has the same name as its API operation’s operationId, and the operation’s summary and description as its title and description: create_code_task is the tool for Create a coding task. It takes the operation’s path, query, and body fields as one set of arguments, returns the same response, and runs the same permission and plan checks. Leave out org_id: the connection already selects the organization. The Access column shows the hint each tool gives your client. Read marks tools that only read data, which your client sees as read-only. Write marks every other tool, including previews that change nothing, such as preview_triage_rule and preview_repository_configuration. Your client can use the hint to ask you before the agent runs a Write tool.

Scopes by area

A tool needs the read scope of its group to read data and the write scope to change it. Every tool except connection_status, list_pull_requests, get_pull_request, start_code_review, and get_code_review also needs account:read, so include it in every scope list. If your connection lacks a scope, the call fails with insufficient_scope and names the missing scope; add it in your client and sign in again, as Choose scopes describes.

Differences from the API

Tools marked MCP only have no API operation with the same name:
  • connection_status checks the connected account, organization, and region.
  • list_pull_requests and get_pull_request find and read pull requests by repository, branch, or pull request URL. They take the place of the List pull requests and Get a pull request operations, which aren’t tools.
  • start_code_review and get_code_review run a CodeRabbit review of local code, which the CodeRabbit CLI uploads from your machine.
  • list_organizations, list_organization_repositories, list_linked_repositories, list_users, manage_seats, and list_audit_logs do the jobs of export and user-management operations such as List users and List audit log entries, in the connected organization only. manage_seats takes user IDs and doesn’t work in an Enterprise SSO workspace; there, manage seats with a workspace API token.
These API operations have no tool, so call them through the API or the CLI:
  • The metrics exports, such as review metrics.
  • The organization Learnings operations on /v1/learnings. The Knowledge Base tools, such as list_learnings, work with Learnings too; see Learnings.
  • The Deep Scan and Bug Scan findings exports.
  • Adding users to a self-hosted organization, changing roles, and changing the seat assignment mode.
  • download_verify_evidence, which returns a file. The get_verify_evidence tool returns the evidence details.

Tools

Organizations and repositories

Pull requests and reviews

Coding Agent

Coding environments

Verify

Triage

Security

Knowledge Base

Automations

Reports

Configuration

Connections

Users and roles

Usage and metrics

What’s next

Connect an MCP client

Connect your agent and choose the scopes for the tools it calls.

Request and read reviews

Find a pull request, request a review, and follow its status.