Turn YepCode processes into MCP tools. No deployment.

Expose real actions to your agents: tag a YepCode process and it shows up as an MCP tool your AI client can call. Define inputs with JSON Schema, run securely in YepCode sandboxes, and choose hosted (always on) or self-hosted (open source).

Quick connect

Hosted

cloud.yepcode.io/mcp

Always on (24/7). Auth via OAuth or API token.

Build tools, not servers

Your deployment is publishing a process in YepCode. The MCP server simply exposes tagged processes as tools, so your agents can act without you running new infrastructure.

Any process becomes a tool

Add a tag (like `mcp-tool`) to a process and it becomes discoverable as an MCP tool in Cursor, Claude, and other clients.

Inputs described with JSON Schema

Give models rich parameter metadata (types, required fields, enums, defaults) so tool calls are accurate and predictable.

Secure, production-ready execution

Tools run inside YepCode's isolated environments with secrets, dependencies, logs, and auditability built in.

Two ways to connect

Use the hosted MCP endpoint for zero ops, or run the open-source server yourself for full control.

Remote — always on

Hosted by YepCode

Hosted MCP endpoint managed by YepCode.

  • Always on 24/7 (hosted)
  • OAuth or API token auth
  • No install, no upgrades, no maintenance

Endpoint

cloud.yepcode.io/mcp

Local — self-hosted

Open source

For air-gapped environments or when you want full control. The MCP server is open source and runs via NPX.

  • Open source server
  • Runs via NPX (or Docker)
  • Best for air-gapped or custom networking

How it works

Create or reuse a YepCode process

Write Python or Node.js code that talks to your APIs, databases, and internal systems.

Define parameters

Add an input schema (JSON Schema) so the MCP client knows exactly how to call your tool.

Tag what you want to expose

Use tags like `mcp-tool`, `core`, or `automation` and expose only the toolsets you choose.

Connect your MCP client

Point Cursor/Claude to the hosted endpoint (always on) or run the open-source server locally.

Built-in tools included

Your own processes become tools, and the server also ships with foundational tools for common agent workflows.

`run_code` for safe execution

Run AI-generated JavaScript or Python in YepCode's secure environment when you need quick glue code.

Storage management tools

Upload, download, list, and delete files to support multi-step agent workflows.

API management tools

Manage processes, executions, schedules, variables, modules, and more via YepCode API tools.

OAuth or API tokens

Choose an auth model that fits your environment, and keep control over who can call what.

Why YepCode for MCP tools?

Dynamic toolsets with tags

Group tools by tags and expose only what you want. Your MCP tools stay in sync as you add, version, and evolve processes.

One MCP server, many runtimes

Implement tools in Python or Node.js (or both) without splitting your toolchain across multiple servers.

Works with any MCP client

Use Cursor, Claude Desktop, or any MCP-compatible platform to discover and call your tools.

Hosted or open source

Start with the always-on hosted endpoint, and self-host the open-source server whenever you need full control.

Ready to connect?

For setup steps, tool reference, and process tags, see the MCP Server docs .

For installation details and source code, see yepcode/mcp-server-js .