MCP Server

Kubit MCP is a Model Context Protocol integration that allows AI development tools like Cursor and Claude to securely access Kubit's analytical capabilities through standardized MCP tools. By connecting an MCP-compatible IDE or assistant to Kubit, users can explore schemas, execute Reports, retrieve raw data and find existing Reports without leaving their development workflow.

Kubit's MCP server exposes several tools that mirror common actions performed in the Kubit platform, enabling a highly productive, code-assisted analytics experience.

What the Kubit MCP Server Provides

The Kubit MCP server includes a set of analytical tools that developers or analysts can use programmatically.

Available MCP Tools

• getUserContext – Initialize User Context
  • Always the first call. Retrieves:
    • User context
    • Available schemas
    • Active schema selection
  • Parameters: Initialization details required by the Kubit application.
• getSchema – Explore Schema Details
  • Retrieves full metadata for a specific schema, including:
    • Event definitions
    • Properties
    • Dimensions
  • Parameters: Reference to the desired schema.
• createReport – Create and Execute Reports
  • Generates analytical queries and returns:
    • A summary of results
    • A link to the full report in the Kubit app
    • A formulaTemplateId and analysisId for further editing
  • Parameters: Arguments required by Kubit to generate a new report.
• getRawData – Export Raw Report Data
  • Exports CSV-like raw output from an existing report.
  • Parameters: Reference to the previously generated report.
• searchKubit – Search Reports and Dashboards
  • Searches for reports or dashboards across your organization.
  • Parameters: Query text suitable for locating Kubit assets.

Connecting Kubit MCP to Cursor

1. Open Customize → MCPs
2. Select New MCP Server.
3. Paste the following JSON configuration:

{
  "mcpServers": {
    "KubitMCP": {
      "url": "https://mcp.kubit.ai/mcp"
    }
  }
}

4. Create a project folder if prompted.

5. Click Connect next to the MCP server.
6. Enter your organization name (org: <org name>).
7. Complete the authentication and authorization steps.

Troubleshoot

• Open Cursor Settings (Cmd/Ctrl + Shift + J), go to "Tools & MCPs" -> "Installed MCP Servers"
• Find KubitMCP, click the icon next to it to Restart (refresh) or Disable/Enable it.
• You may also consider reload Cursor by completely reload the editor context: Open the command palette (Cmd/Ctrl + Shift + P) and run Developer: Reload Window.

Connecting Kubit MCP to Claude Desktop

1. Open Settings → Connectors.
2. Select Add custom connector.
3. Name the connector KubitMCP.
4. For Remote MCP URL, enter: https://mcp.kubit.ai/mcp
5. Click Add.
6. Enter your organization name (org: <org name>).
7. Complete the Kubit login and authorization process.

Troubleshoot

• Use /mcp command to check the status of the KubitMCP server.
• If the connector is listed but no tools appear, open Settings → Connectors, disable and re-enable it.
• Changing the Remote MCP URL invalidates the stored credentials. Re-authorize the connector after any URL change.

Connecting Kubit MCP to Claude Code

1. Add the server from your terminal:

claude mcp add --transport http KubitMCP https://mcp.kubit.ai/mcp

2. Run /mcp inside Claude Code, select KubitMCP, then choose Authenticate. You can also run claude mcp login KubitMCP.
3. Enter your organization name (org: <org name>).
4. Complete the Kubit login and authorization process.

By default the server is added at local scope — your user, current project only. Add --scope user to make it available in all your projects, or --scope project to share it with your team through version control. Project scope writes a .mcp.json file at the project root:

{
  "mcpServers": {
    "KubitMCP": {
      "type": "http",
      "url": "https://mcp.kubit.ai/mcp"
    }
  }
}

Troubleshoot

• Use /mcp command to check the status of the KubitMCP server.
• Run claude mcp list to confirm the server is configured and connected, or claude mcp get KubitMCP for its details.
• If the server connects but tool calls fail, the stored token is stale. Run claude mcp logout KubitMCP, then claude mcp login KubitMCP.
• Reports over wide date ranges may exceed the default tool timeout. Narrow the date range if calls time out.

Connecting Kubit MCP to Codex

1. Add the server from your terminal:

codex mcp add KubitMCP --url https://mcp.kubit.ai/mcp

Alternatively, add it directly to ~/.codex/config.toml:

[mcp_servers.KubitMCP]
url = "https://mcp.kubit.ai/mcp"
auth = "oauth"
startup_timeout_sec = 30
tool_timeout_sec = 120

2. Authenticate:

codex mcp login KubitMCP

3. Enter your organization name (org: <org name>).
4. Complete the Kubit login and authorization process.

For a project-scoped setup, create .codex/config.toml in the project root with the same block. Codex merges project configuration over your global file.

Troubleshoot

• Run codex mcp list to confirm the server is configured and connected.
• If the server never appears, your Codex build may keep remote MCP support behind a feature flag, in which case a URL-based server is skipped silently. Add the following to ~/.codex/config.toml:

[features]
experimental_use_rmcp_client = true

Newer builds rename this to rmcp_client and recent versions enable it by default. Upgrading Codex with npm install -g @openai/codex is the cleaner fix.

• If the server connects but tool calls fail, the stored token is stale. Run codex mcp login KubitMCP again.
• If the browser callback page shows a connection error, the local listener has expired. Start the login again and complete it without pausing.

Additional Reference: GitHub Repository

For developers who want to review the MCP server source code, configuration patterns, or contribution guidelines, the full repository is available here: https://github.com/Kubit-AI/mcp-server