YepCode MCP Server
Execute LLM-generated code in secure YepCode sandboxes and expose YepCode processes as callable MCP tools for AI assistants.
YepCode MCP Server connects AI assistants to YepCode Cloud, a serverless runtime for running scripts and automations. It lets agents execute arbitrary LLM-generated Node.js or Python code in isolated, production-ready sandboxes and turns your existing YepCode processes into individual MCP tools that agents can call directly.
The server exposes a run_code tool for ad-hoc code execution along with management tools for environment variables, storage files, processes, schedules, modules, and executions. Tagged YepCode processes are automatically surfaced as MCP tools with their parameter schemas, making it easy to expose curated, governed automations to an agent. Tool exposure can be tuned via YEPCODE_MCP_TOOLS to mix and match categories like run_code, yc_api, yc_api_full, core, or automation.
YepCode maintains the project as the official integration (org yepcode on GitHub) and also offers a fully hosted MCP endpoint at https://cloud.yepcode.io/mcp so users can skip local installation. The repo is MIT licensed and ships both NPX and Docker deployment options.
Tools
| Tool | Description |
|---|---|
run_code |
Executes code (Node.js or Python) in a secure YepCode sandbox with optional language and runtime settings. Returns execution results, console output, and error messages. |
set_env_var |
Creates or updates a workspace environment variable, with optional log masking. |
remove_env_var |
Deletes a workspace environment variable. |
list_files |
Lists files in YepCode storage, with optional prefix filtering. |
upload_file |
Uploads a file to YepCode storage. Supports base64 encoding for binary content. |
download_file |
Downloads a file from YepCode storage. |
delete_file |
Deletes a file from YepCode storage. |
get_processes / create_process / get_process / update_process / delete_process |
CRUD operations for YepCode processes. |
execute_process_async / execute_process_sync |
Run a YepCode process synchronously or asynchronously with input parameters. |
schedule_process |
Schedule a YepCode process to run on a cron-like schedule. |
get_schedules / get_schedule / pause_schedule / resume_schedule / update_schedule / delete_schedule |
Manage existing process schedules. |
get_variables / create_variable / update_variable / delete_variable |
Manage workspace variables via the YepCode API. |
get_storage_objects / upload_storage_object / download_storage_object / delete_storage_object |
Manage YepCode storage objects via the API. |
get_executions / get_execution / kill_execution / rerun_execution / get_execution_logs |
Inspect, terminate, rerun, and read logs for process executions. |
get_modules / create_module / get_module / delete_module |
Manage reusable code modules. |
get_process_versions / get_module_versions / get_module_version / delete_module_version / get_module_aliases |
Extended version management tools available when using the yc_api_full category. |
Prerequisites
- A YepCode Cloud account (sign up at https://cloud.yepcode.io)
- An API token from Settings > API credentials in your YepCode workspace
- Node.js 18+ if using the NPX install method
Option 1: NPX (local)
Add to your MCP client config (Claude Desktop, Cursor, etc.):
{
"mcpServers": {
"yepcode-mcp-server": {
"command": "npx",
"args": ["-y", "@yepcode/mcp-server"],
"env": {
"YEPCODE_API_TOKEN": "your_api_token_here"
}
}
}
}
Option 2: Docker
{
"mcpServers": {
"yepcode-mcp-server": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"YEPCODE_API_TOKEN=your_api_token_here",
"yepcode/mcp-server"
]
}
}
}
Option 3: Hosted MCP endpoint
Skip local installation and point your MCP client at the YepCode-hosted server:
https://cloud.yepcode.io/mcp
Configuration
Optional environment variables:
YEPCODE_MCP_OPTIONS: comma-separated flags such asrunCodeCleanup(preserve source after execution) orskipCodingRules(omit coding guidelines from tool schema).YEPCODE_MCP_TOOLS: comma-separated tool categories to expose, for examplerun_code,yc_api,run_code,yc_api_full, orcore,automation.
- Give an agent a sandbox to write and execute Node.js or Python on the fly for data transforms, scraping, or one-off computations without touching local infrastructure.
- Expose vetted YepCode processes (e.g. a Stripe refund flow or a HubSpot enrichment script) as named tools an LLM can safely call with structured inputs.
- Schedule, pause, resume, or rerun YepCode processes from an AI chat to operate background automations.
- Inspect execution logs, kill stuck runs, and rerun failed jobs as part of an agent's incident response workflow.
- Manage workspace environment variables and storage files (upload/download/delete) for use inside automations.
- "Run a Python snippet that fetches https://example.com/data.json and returns the top 5 records by revenue."
- "Execute my
sync-hubspot-contactsYepCode process for tenant 42 and show the logs when it finishes." - "Pause the nightly
cleanup-orphansschedule and resume it tomorrow at 9am UTC." - "Upload this CSV to YepCode storage as
leads-2026-05.csvand then run theimport-leadsprocess against it." - "Find the last failed execution of
send-weekly-reportand rerun it."
- Official, actively maintained by YepCode (org-owned repo with recent releases).
- Broad tool coverage: ad-hoc code execution plus full CRUD over processes, schedules, modules, storage, and executions.
- Multiple deployment options: NPX, Docker, or a fully hosted MCP endpoint at
cloud.yepcode.io/mcp. - Polyglot sandbox supports Node.js and Python with isolated, production-ready runtimes.
- Requires a YepCode Cloud account and API token; tied to YepCode's hosted runtime rather than running code locally.
- Exposing the full
yc_api_fulltoolset can flood an agent's context with many similar tools, so you usually need to tuneYEPCODE_MCP_TOOLS. - Documentation lives mostly in the README and YepCode docs site; deeper examples per tool are limited.
- E2B MCP Server for sandboxed code execution via E2B's runtime.
- Pipedream MCP for exposing workflows and 2,000+ app integrations as MCP tools.
- Cloudflare Workers / Sandbox MCP for running code in Cloudflare's serverless environment.