Make

What the Make MCP server does

The Make MCP server supports AI-assisted work with Make automation scenarios. Its associated skills cover account orientation, scenario explanation, scenario creation and editing, activation and execution review, and debugging. The material also includes a separate Endpoints skill for direct app actions through Make’s @makehq/endpoints-sdk package or its command-line interface.

The scenario skills are organized by task. Exploration covers organizations, teams, scenario listings, connection health, and account-level checks. Building covers app and module discovery, connections, account-specific values, flow control, error handling, subscenarios, and AI agents. Operations covers running scenarios, changing activation state, reviewing executions, debugging, and investigating webhooks.

How it works

Scenario skills use Make’s scenario-management MCP server over HTTP. The repository provides platform-specific URLs for Claude Code, Cursor, and OpenAI Codex, and other agents can use the matching URL in their JSON MCP configuration. On first use, authentication goes through Make’s OAuth consent screen.

The Make MCP server requests one permission bundle. All requested permissions must be granted for the server to work; reconnecting and granting the complete set is the documented response to permission errors or 403 responses. The reference skill explains shared concepts such as scopes, the data and remark contract, structure versus configuration, the one-save write model, and operations the tool surface will not author.

The Endpoints skill is separate from the scenario MCP interface. It uses the Endpoints SDK or CLI for discovery, managed connections, typed calls, complete retrieval, and approval-bound writes with readback. Endpoints are described as being in closed beta, and installing the skill does not install the package or grant beta access.

Setup and configuration

A Make account is required. For scenario work, the account also needs access to Make’s scenario-management MCP server. Install the skills through the repository’s plugin options or download the skill archives into an agent’s skills directory. The repository documents directories for Codex, Claude Code, Cursor, Windsurf, Cline, and generic agents.

If a plugin does not register the server, add the appropriate HTTP endpoint manually. Claude Code uses https://mcp.make.com/claude, Cursor uses https://mcp.make.com/cursor, and OpenAI Codex uses https://mcp.make.com/openai. Endpoints-only use does not require the scenario MCP server, but it does require an approved Node.js or terminal environment, the Endpoints package, and organization access to the beta.

Tools and capabilities

• Explore organizations, teams, scenarios, connections, and account-wide status.
• Explain existing scenarios and their behavior.
• Discover apps and modules while building or editing scenarios.
• Resolve account-dependent configuration, flow control, error handling, subscenarios, and AI-agent setups.
• Run, activate, deactivate, review, and debug scenarios.
• Investigate webhook executions.
• Discover app Endpoints and make typed SDK or CLI calls with managed connections.
• Perform approval-bound Endpoint writes and retrieve results for readback.

Limitations and notes

The Make MCP server does not author every capability supported by the Make editor. When a tool refuses an operation, the documented approach is to inspect the refusal and current tool contract, then use the Make editor for explicitly unsupported work.

The Make MCP server requires network access to Make services and complete OAuth permissions. Endpoints have separate requirements for the selected key or zone, Make scopes, team access, provider connections, and organization beta or runtime access. Local catalog discovery and package installation do not enable Endpoint execution.