MCP Server

The DebugBundle MCP server exposes debugging tools that AI agents (Codex, Claude Code, GitHub Copilot, Cursor, Cline, and others) can call directly. Instead of switching between your IDE and a dashboard, your agent can query incidents, retrieve bundles, inspect aggregate product usage and AnalyticsBundles, manage tokens, and run reproductions inline.

What is MCP?

The Model Context Protocol is an open standard for connecting AI assistants to external tools. MCP servers expose typed tool definitions that agents discover and invoke during conversations.

DebugBundle implements an MCP server that wraps the same domain services used by the API and CLI, ensuring consistent behavior across all interfaces.

Supported Node.js versions: 22.x through 26.x.

Official Registry

Available on the official MCP Registry as com.debugbundle/mcp.

• Registry listing: com.debugbundle/mcp
• API record: /v0.1/servers/com.debugbundle%2Fmcp/versions/latest
• Distribution status and marketplace listings: MCP Distribution

Install and Use

Run the MCP server directly with npm:

npx @debugbundle/mcp

Or install it globally:

npm install -g @debugbundle/mcp
debugbundle-mcp

The package ships MCP Registry metadata as server.json with server name com.debugbundle/mcp.

The supported general-purpose MCP path remains the local stdio package. A separate read-only OpenAI Plugin candidate now serves its OAuth-protected production endpoint at https://mcp.debugbundle.com/mcp; it is not submitted, published, or publicly installable yet.

Install Matrix

Authentication

The following modes apply to the local stdio package. The separate OpenAI candidate uses OAuth/OIDC, PKCE S256, explicit read-only resource scopes, and user consent; it never accepts a project token or per-tool bearer credential.

The MCP server supports three authentication modes:

When running locally alongside the CLI, the MCP server automatically picks up your authenticated session. No additional configuration is required.

If the CLI is not authenticated yet, bootstrap it first with any of these paths:

debugbundle login
debugbundle login dbundle_mem_xxxxxxxxxxxx
debugbundle login --github
debugbundle login --github-device

debugbundle login with no flags opens an interactive chooser in a TTY. --github first tries an existing gh auth token for a fully headless agent flow, then falls back to GitHub device flow when browser approval is needed. Email-code signup and manual token creation remain available in the web app for users who do not use GitHub.

For non-OpenAI remote, marketplace-managed, or headless environments, prefer setting the member token in the MCP server environment:

{
  "mcpServers": {
    "debugbundle": {
      "command": "npx",
      "args": ["@debugbundle/mcp"],
      "env": {
        "DEBUGBUNDLE_MEMBER_TOKEN": "dbundle_mem_a1b2c3d4..."
      }
    }
  }
}

Use DEBUGBUNDLE_API_URL in the same environment block only for self-hosted or non-production API hosts. Individual tool calls can still pass a member token as bearerToken:

{
  "tool": "list_incidents",
  "arguments": {
    "bearerToken": "dbundle_mem_a1b2c3d4...",
    "limit": 5
  }
}

Configuration

Codex

Use the dedicated Codex guide for the developer plugin or direct MCP configuration, local and hosted authentication, verification, updates, and removal. Codex developer support uses the same stdio server as other coding agents and does not wait for public OpenAI plugin approval.

Claude Desktop

Open Claude Desktop Settings > Developer, edit the local MCP configuration, and add:

{
  "mcpServers": {
    "debugbundle": {
      "command": "npx",
      "args": ["-y", "@debugbundle/mcp"]
    }
  }
}

Restart Claude Desktop after saving. The local server can reuse debugbundle login auth state on the same machine. For managed/headless setups, set DEBUGBUNDLE_MEMBER_TOKEN in the MCP server environment. Set DEBUGBUNDLE_API_URL only for self-hosted or non-default API hosts.

Claude Code Plugin

Use the dedicated Claude Code guide for the repository plugin or direct MCP setup, authentication, verification, updates, and removal. Both documented paths use the local-auth profile.

VS Code (GitHub Copilot)

Add to your VS Code settings or .vscode/mcp.json:

{
  "mcp": {
    "servers": {
      "debugbundle": {
        "command": "npx",
        "args": ["@debugbundle/mcp"]
      }
    }
  }
}

Cursor

Add to your Cursor MCP configuration:

{
  "mcpServers": {
    "debugbundle": {
      "command": "npx",
      "args": ["@debugbundle/mcp"]
    }
  }
}

Generic MCP Client

Any MCP-compatible client can connect over stdio with the published package:

npx @debugbundle/mcp

The server communicates over stdin/stdout using the MCP JSON-RPC protocol.

Troubleshooting

Tool Categories

The MCP server provides 117 tools organized into these categories:

See the MCP Tools Reference for the complete tool catalog with parameters, and MCP Workflows for practical agent workflow patterns.

Next Steps

• OpenAI Plugin — Candidate scope, privacy boundaries, and release evidence
• MCP Distribution — Official package, registry, marketplace, and skill channels
• MCP Tools Reference — Complete tool catalog
• MCP Workflows — Agent workflow patterns
• CLI Cloud Workflow — CLI authentication (shared with MCP)