MCP Sidecar
Start, monitor, and tear down long-running dev servers, workers, and scripts from any MCP client without losing stdout/stderr context.
Overview
io.github.lsequeiraa/mcp-sidecar is a MCP server for the Build phase that manages long-lived background processes and their output for any MCP client over stdio.
What is this MCP server?
- Cross-platform stdio MCP server mcp-sidecar v0.3.2 (npm).
- Configurable SIDECAR_MAX_PROCESSES (default 10 concurrent jobs).
- Per-stream SIDECAR_BUFFER_SIZE default 1048576 bytes and optional SIDECAR_MAX_OUTPUT_SIZE cap on output tool.
- Graceful shutdown with SIDECAR_KILL_TIMEOUT (default 5000 ms) and SIDECAR_CLEANUP_AFTER auto-remove (default 1800 s).
- Server version 0.3.2 on npm identifier mcp-sidecar
- Default SIDECAR_MAX_PROCESSES: 10
- Default SIDECAR_BUFFER_SIZE: 1048576 bytes per stream
What problem does it solve?
Coding agents lose track of dev servers and workers they start, making debugging and shutdown messy during long agent sessions.
Who is it for?
Solo builders using Claude Code or Cursor who want the agent to run and supervise local servers and scripts safely.
Skip if: Production orchestration on Kubernetes or teams that only need single inline terminal commands without lifecycle management.
What do I get? / Deliverables
After adding the sidecar MCP server, agents get named processes with buffered logs, graceful kill, and automatic cleanup of exited jobs.
- Managed background processes with retrievable stdout/stderr via MCP tools
- Bounded concurrency and cleanup of exited processes on the dev machine
Recommended MCP Servers
Journey fit
Agent tooling in Build is where you give models safe control surfaces over local processes instead of one-shot shell guesses. Agent-tooling fits a dedicated sidecar that manages background processes across platforms for Claude Code-style loops.
How it compares
Local process sidecar MCP, not a cloud deploy platform or a generic bash-execution skill.
Common Questions / FAQ
Who is io.github.lsequeiraa/mcp-sidecar for?
Indie developers and agent users who need reliable start/stop/log access to background dev processes from MCP clients.
When should I use io.github.lsequeiraa/mcp-sidecar?
Use it during active Build sessions when the agent should keep APIs, bundlers, or workers running while reading their output over multiple turns.
How do I add io.github.lsequeiraa/mcp-sidecar to my agent?
Install npm package mcp-sidecar, add it as a stdio MCP server in Claude Code or Cursor, and optionally set SIDECAR_MAX_PROCESSES and buffer env vars.