MCP Server

The MCP server is currently in beta. We’d love your feedback at contact@cubic.dev.

When your AI coding assistant tries to understand your codebase to build a feature, it often has to do significant work to figure out how things connect — and may not do it as thoroughly as a dedicated analysis would. cubic’s AI wiki solves this by pre-computing that understanding — spending significant reasoning and tokens to analyze your codebase and distill it into structured documentation. The MCP server gives AI assistants like Cursor, Claude, Windsurf, and OpenCode direct access to this cached knowledge — so they can query pre-synthesized context instead of figuring it out from scratch. With auto-refresh enabled, this stays up to date as your codebase evolves.

Integrations page showing MCP Server card in Settings

What you can do

Once connected, your AI assistant can:

Access AI Wikis - Query AI-generated codebase documentation containing architecture overviews, feature explanations, and code references
Review codebase scans - List scans, inspect issues, and pull detailed findings with code context
Inspect AI review learnings - Understand team-specific review patterns learned from feedback and scans
Explore repositories - Discover and navigate wiki pages for any repository in your installation
Understand codebases - Get contextual information about how systems work, design patterns, and component relationships

The MCP server requires a cubic account with access to at least one repository. You must be a member of the repository’s installation to access its wiki, scans, and learnings.

Use cases

The MCP server supports different workflows depending on your needs:

Local development

Connect your AI coding assistant (Cursor, Claude Desktop, Windsurf) to cubic for enhanced codebase understanding. Each developer generates their own API key and stores it in their personal config file — nothing needs to be committed to the repository.

Internal tooling

Build internal applications that need access to codebase documentation. For example:

Customer support tools that query wiki pages to help support engineers understand product behavior
Onboarding systems that surface relevant architecture docs for new team members
Custom chatbots that answer questions about your codebase

For internal tooling, store the API key as an environment variable or in your secrets manager.

Quick setup

Generate an API key

Go to Settings → Integrations → MCP Configuration and click Generate API Key. Copy the key immediately—it’s only shown once.

API keys are personal to each user. Each team member generates their own key, which grants access only to repositories they’re authorized to view. Don’t share keys between team members.

MCP Server configuration showing API key and client setup

Configure your AI client

Add the cubic MCP server to your preferred AI coding assistant using the configuration below.

Start using MCP tools

Ask your AI assistant to list wikis, inspect scan findings, or retrieve review learnings for your repositories.

Client configuration

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "cubic": {
      "url": "https://www.cubic.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer cbk_your_api_key_here"
      }
    }
  }
}

Restart Cursor after adding the configuration for changes to take effect.

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "cubic": {
      "url": "https://www.cubic.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer cbk_your_api_key_here"
      }
    }
  }
}

Run the following command to add the cubic MCP server:

claude mcp add --transport http cubic https://www.cubic.dev/api/mcp --header "Authorization: Bearer cbk_your_api_key_here"

Then run /mcp in a Claude Code session to verify the connection.

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "cubic": {
      "url": "https://www.cubic.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer cbk_your_api_key_here"
      }
    }
  }
}

Add to ./opencode.json in your project root:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "cubic": {
      "type": "remote",
      "url": "https://www.cubic.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer cbk_your_api_key_here"
      }
    }
  }
}

OpenCode uses a per-project configuration format rather than a global config file.

Press Ctrl/Cmd + Shift + P and search for MCP: Add Server 2. Select HTTP as the transport type 3. Enter the URL: https://www.cubic.dev/api/mcp 4. Add the authorization header when prompted

For MCP clients not listed above, use these connection details:

Setting	Value
URL	`https://www.cubic.dev/api/mcp`
Transport	HTTP (Streamable)
Authentication	Bearer token in `Authorization` header

Available tools

Wiki tools

cubic AI Wikis are AI-generated codebase documentation containing architecture overviews, feature explanations, code references, and diagrams. Use these tools to explore and understand codebases.

Enable auto-refresh on your wiki to keep documentation up to date as your codebase evolves.

Tool	Description	Parameters
`list_wikis`	List all wikis you have access to	`search` (optional)
`list_wiki_pages`	List all available wiki pages for a repository	`owner`, `repo`
`get_wiki_page`	Retrieve the content of a specific wiki page	`owner`, `repo`, `pageRef`

Recommended workflow:

Call list_wikis to discover repositories with generated wikis
Call list_wiki_pages to discover available documentation for a repository
Use get_wiki_page with a page reference to read specific content

Example: Exploring a repository's wiki

Ask your AI assistant:

“List the wiki pages for the acme/backend repository”

The assistant will call list_wiki_pages and return something like:

Available wiki pages for acme/backend:
- architecture-overview
- authentication-system
- database-schema
- api-endpoints
- deployment-guide

Then ask:

“Show me the authentication system documentation”

The assistant will call get_wiki_page with pageRef: "authentication-system" and return the full documentation.

Codebase scan tools

cubic codebase scans identify security vulnerabilities, data integrity issues, business logic bugs, and stability concerns.

Tool	Description	Parameters
`list_scans`	List repository scan summaries	`search?`, `limit?`, `offset?`
`get_scan`	List aggregated issues with filters/pagination	`owner`, `repo`, `category?`, `triageStatus?`, `minSeverity?`, `filePath?`, `limit?`, `offset?`
`get_issue`	Get full issue details including code context	`issueId`

Recommended workflow:

Call list_scans to discover repositories with open issue counts and latest run status
Call get_scan with owner/repo and optional filters to list issues
Call get_issue for complete details on a specific issue

You do not need to pass limit for list_scans. Calling list_scans with no parameters returns the default first page (limit=10). list_scans accepts only search?, limit?, and offset?. Deprecated owner/repo inputs are rejected. list_scans returns JSON with repository summaries:

repositories: array of { repositoryId, owner, repo, openIssueCount, lastRunAt, lastRunStatus }
totalCount, hasMore, limit, offset, search

get_scan returns JSON with this envelope:

items: aggregated issue list for the requested repository
totalCount: total filtered issues before visibility limits
visibleCount: issues visible to the caller after plan visibility limits
hasMore, limit, offset
appliedFilters, aggregationMeta

get_scan filter semantics:

minSeverity is inclusive and must be 0..10
filePath is a case-sensitive prefix match
Default limit is 20; server cap is 100

Examples:

list_scans({})
list_scans({ search: "cubic" })
list_scans({ search: "mrge", limit: 5, offset: 5 }) (optional pagination)

Learning tools

cubic AI review learnings are team-specific code review patterns learned from senior reviewer feedback, user corrections, and codebase scans.

Tool	Description	Parameters
`list_learnings`	List all AI review learnings for a repository	`owner`, `repo`
`get_learning`	Get full details of a specific learning	`learningId`

Recommended workflow:

Call list_learnings to discover patterns and preferences for a repository
Call get_learning with a learning ID for full details and original feedback context

Troubleshooting

Connection refused or timeout errors

Verify that:

Your API key is correct and hasn’t been revoked
You’re using the correct endpoint: https://www.cubic.dev/api/mcp
Your firewall or network allows outbound HTTPS connections

401 Unauthorized errors

Check that your API key is prefixed with cbk_ - Ensure the Authorization header uses Bearer format - Regenerate your API key if issues persist

No wiki pages found

Confirm the repository has an AI Wiki generated - Verify you have access to the repository’s installation - Check that the owner and repo parameters match the GitHub repository exactly

MCP client not recognizing the server

Restart your AI client after updating the configuration
Check the config file path is correct for your operating system
Validate your JSON syntax (no trailing commas, proper quoting)

What’s next?

We’re considering adding more tools to the MCP server:

PR review tools - Access AI review comments and suggestions
Codebase search - Search across your indexed repositories
Analytics access - Query review metrics and team insights

Interested in any of these? Let us know — your feedback helps us prioritize.

Get started

AI code review

Development environment

Analytics

Portal

Account & support

What you can do

Use cases

Local development

Internal tooling

Quick setup

Client configuration

Available tools

Wiki tools

Codebase scan tools

Learning tools

Troubleshooting

What’s next?

Get started

AI code review

Development environment

Analytics

Portal

Account & support

​What you can do

​Use cases

​Local development

​Internal tooling

​Quick setup

​Client configuration

​Available tools

​Wiki tools

​Codebase scan tools

​Learning tools

​Troubleshooting

​What’s next?

What you can do

Use cases

Local development

Internal tooling

Quick setup

Client configuration

Available tools

Wiki tools

Codebase scan tools

Learning tools

Troubleshooting

What’s next?