Running a dev server through an ordinary shell call either blocks or dies with the call. Here it is named, kept alive, and its log is on disk — so 'why did the build fail' becomes a grep instead of a re-run.
A process manager exposed over MCP. Long-running bash commands are started under a name, keep running while the conversation continues, and their stdout and stderr are written to disk — so the log outlives the process and can be searched later. A CLI ships alongside for the same operations.
- A named long-running bash command started, with a description and a working directory
- Every process listed, running and terminated, with its state
- Logs retrieved, with the number of recent lines you ask for
- Logs searched by regular expression
- Input sent to a running process, as if typed at it
- One process terminated, or all of them at once
- Finished processes cleared and their logs deleted
- A failure reported immediately when a process does not start
Node — npm install -g pm-mcp, or run it through npx. No account, no key. Commands run in your own shell with your permissions, so give it the same trust you would give a terminal.
One command — npx pm-mcp
