persistproc

persistproc

MCP server + command line tool that allows agents to see & control long-running processes like web servers.

6
Stars
1
Forks
3
Releases

Overview

Persistproc is an MCP server and command line tool that lets agents and humans see and control long-running processes such as web servers. Its goal is to reduce copying and pasting when coding with AI, simplify multi-agent workflows, and remain tool-agnostic. There is no separate config file; processes are managed entirely at runtime, and persistproc is not a replacement for supervisord. The server exposes an MCP API at http://127.0.0.1:8947, allowing clients to list processes, start/stop/restart them, and retrieve their outputs. The distribution ships with a minimal toolset: ctrl for unified process control (start, stop, restart); list to enumerate managed processes with optional filters (PID, command, working directory) and log paths; and output to fetch captured logs. You can start a process by running persistproc npm run dev from your project directory; the command is sent to the server and streaming output is returned in your terminal. The design prioritizes a lightweight, runtime-driven approach suitable for agent-assisted development workflows.

Details

Owner
irskep
Language
Python
License
MIT License
Updated
2025-12-07

Features

Unified process control

ctrl provides start, stop, and restart capabilities for managed processes.

Process listing and status

list enumerates all managed processes, shows their status, and can filter by pid, command, or working directory while exposing log paths.

Output streaming and retrieval

output streams capture from processes and lets clients retrieve logs.

MCP server interface

Exposes an MCP-compatible server at http://127.0.0.1:8947 for programmatic control by clients.

Runtime-managed (no config file)

Processes are managed entirely at runtime with no separate config file.

Concise toolset for AI workflows

A deliberately small set of tools to keep LLMs focused and reduce tool overhead.

Audience

DevelopersRun and supervise long-running dev servers; share outputs with agents.
LLM agentsInteract with processes via MCP to read logs and restart processes.
Human operatorsProvide humans a CLI and MCP interface to control processes and view outputs.

Tags

MCPserverpersistprocprocess managementlogsCLImulti-agentAI agentsstreaming output