Most servers ship a fixed tool list; here it is whichever processes you tagged, each with a schema you wrote, Python and Node.js side by side. run_code covers the rest: when a job needs glue nobody built a process for, the model writes it and YepCode runs it where your secrets already live.
YepCode's server, which exposes the processes in your workspace as callable tools and adds a sandbox where an agent can execute JavaScript or Python it just wrote.
- run_code, which executes JavaScript or Python in YepCode's isolated environment and returns the value, the logs, or the error
- Your own processes as tools: tag one mcp-tool and it appears immediately, or use custom tags and list them in YEPCODE_MCP_TOOLS
- A JSON Schema per process, so the model gets types, required fields, enums and defaults instead of guessing at inputs
- Storage tools — list_files, upload_file, download_file, delete_file — handling text and base64 binary across a multi-step task
- Environment variables through set_env_var and remove_env_var, sensitive by default and masked in logs
- API management through yc_api, or yc_api_full for version management, covering processes, schedules, variables, storage, executions and modules
- Processes run synchronously or asynchronously, returning an execution id either way
- Python and Node.js tools living in the same server rather than split across two
A YepCode account. The hosted endpoint is https://cloud.yepcode.io/mcp and takes either OAuth or a token from Settings → API Credentials, passed in the URL or as an Authorization: Bearer header. Self-hosting needs Node.js 18 or later and a YEPCODE_API_TOKEN, run through NPX or Docker.
One command plus a key — npx -y @yepcode/mcp-server, then supply credentials