PRIMS

What hileamlakB/PRIMS MCP server does

hileamlakB/PRIMS MCP server gives MCP-compatible agents access to short-lived Python execution environments. The main execution tool accepts Python source code and can install requested pip packages for that invocation. Standard output and standard error are returned to the caller.

Each execution is designed to be isolated and disposable. This makes the server suitable for temporary calculations, data inspection, transformations, and scripts that do not need state to survive between sessions.

The server is categorized as an 👨‍💻 Code Execution MCP server. It is a fit for agents that need to run Python rather than only generate code, provided the agent can supply the required inputs and handle any resulting files.

How it works

PRIMS can run locally with Python or in Docker. The local startup path uses the repository setup script to create a .venv and install dependencies. The documented server command is python -m server.main, which exposes an MCP endpoint at http://0.0.0.0:9000/mcp. A Docker helper script builds and starts the container and prints the MCP URL when ready.

A session workspace holds files created during execution. mount_file downloads a remote file once per session into the mounts/ path, allowing later Python code to reuse it without sending the URL again. list_dir shows workspace entries, while preview_file returns up to 8 KB from a text file for inspection.

Files created under output/ can be sent to permanent storage with persist_artifact. That tool requires a client-provided presigned URL. Small artifacts may also be retrieved from the server’s artifact endpoint using the MCP session ID.

Setup and configuration

For local development, make scripts/setup_env.sh executable, run it, and activate the resulting virtual environment in each new shell. Start the service with python -m server.main. Alternatively, run scripts/docker_run.sh after making it executable.

Presigned URLs and remote file URLs are supplied as tool arguments rather than documented environment variables.

Tools and capabilities

• run_code: Execute Python with optional pip dependencies and return stdout or stderr.
• list_dir: List files and directories in the current session workspace.
• preview_file: Read up to 8 KB of a text file.
• mount_file: Download a remote file once per session into mounts/<path>.
• persist_artifact: Upload a file from output/ to a presigned URL.

The examples cover listing tools, running code, mounting and reusing a CSV, inspecting a workspace, and persisting a generated artifact. hileamlakB/PRIMS MCP server does not document broader external service integrations beyond the presigned upload and remote-file download flows.

Limitations and notes

The project is marked alpha in the README. Treat arbitrary code execution as an operational risk and evaluate the deployment environment before exposing the endpoint to untrusted users.

Workspace persistence is temporary by design. Files that must survive a session need to be uploaded through persist_artifact or otherwise retrieved before cleanup. preview_file is intended for quick text inspection and is limited to 8 KB. The server’s documented transports are MCP/stdio and the HTTP endpoint described above.