bg-tasks-mcp
MCP server that enables AI agents to start, monitor, and stop background processes like dev servers, builds, and tests, with a web dashboard.
README
bg-tasks-mcp
Let your AI agent manage background processes. One sentence to start, monitor, and stop dev servers, builds, and tests — without leaving the conversation.
Why
Working on a full-stack project means juggling multiple terminal windows to start dev servers, check logs, and kill processes. Every context switch breaks your flow. You're the process manager, not the AI.
bg-tasks flips this: the LLM manages background processes for you.
Before:
┌──────────┐ ┌──────────────┐ ┌──────────────┐
│ opencode │ │ Terminal #1 │ │ Terminal #2 │
│ write code│ │ npm run dev │ │ go run . │
│ ask Q │ │ check logs │ │ check logs │
└──────────┘ └──────────────┘ └──────────────┘
↑ Alt+Tab, manual PID hunting, context switching
After:
┌──────────────────────────────────────────────────┐
│ opencode │
│ │
│ "Start frontend and backend with bg-tasks" │
│ → bg_start("frontend", "npm run dev") │
│ → bg_start("backend", "go run .") │
│ │
│ "Show me frontend logs" │
│ → bg_logs(id="frontend", tail=20) │
│ │
│ "Stop the backend" │
│ → bg_kill(id="backend") │
│ │
│ Dashboard: http://127.0.0.1:9876 │
└──────────────────────────────────────────────────┘
Use Cases
| Scenario | Example |
|---|---|
| Dev servers | bg_start("fe", "npm run dev") + bg_start("be", "python app.py") |
| Builds | bg_start("build", "npm run build") → bg_wait("build") → bg_logs("build") |
| Tests | bg_start("test", "npm test") → bg_wait("test") |
| Long scripts | bg_start("seed", "node seed-db.js") — check progress with bg_logs |
| Multi-project | Open two opencode windows → dashboard shows both projects' tasks |
Installation
npm (recommended)
{"command": ["npx", "-y", "bg-tasks-mcp"]}
No install needed — npx downloads and runs on first use.
Manual install
npm install -g bg-tasks-mcp
Then configure:
{"command": ["bg-tasks-mcp"]}
Git clone
git clone https://github.com/rookie136/bg-tasks-mcp.git
cd bg-tasks-mcp
npm install
Then:
{"command": ["node", "/path/to/bg-tasks-mcp/index.js"]}
When you need to start, monitor, or stop background processes (dev servers, builds, tests),
use the bg-tasks tools.
Usage
6 MCP Tools
| Tool | Purpose | Example |
|---|---|---|
bg_start |
Start a background task | bg_start(name="frontend", command="npm run dev") |
bg_wait |
Wait for completion or timeout | bg_wait(id="frontend", timeout=30) |
bg_logs |
Read task output | bg_logs(id="frontend", tail=20, stream="stdout") |
bg_status |
Inspect or list tasks | bg_status(id="frontend") / bg_status() |
bg_send |
Send text or signal | bg_send(id="frontend", input="<Enter>") |
bg_kill |
Terminate a task | bg_kill(id="frontend") / bg_kill(id="frontend", force=true) |
5 MCP Resources
| Resource | Description |
|---|---|
bg://status |
All tasks as JSON |
bg://{id}/status |
Single task detail + log summary |
bg://{id}/logs |
Combined stdout/stderr |
bg://{id}/logs/stdout |
stdout only |
bg://{id}/logs/stderr |
stderr only |
Web Dashboard
When the first opencode window starts, the HTTP dashboard comes online automatically:
http://127.0.0.1:9876 English (default)
http://127.0.0.1:9876/?lang=zh 中文
Features:
- Window list (owner PID, remote PIDs, task counts, last active time)
- Real-time running tasks view
- Clickable log links for remote window tasks
- Per-window history with clear buttons
- Bilingual UI (EN/中文 toggle)
Architecture
Multi-Window Design
┌─ opencode window #1 (owner) ──────────────────────────────────────┐
│ bg-tasks process │
│ │
│ serveStdio (MCP stdio) ←─ LLM tool calls │
│ HTTP :9876 │
│ GET / → dashboard HTML (bilingual) │
│ GET /status → JSON API │
│ GET /logs/{id} → log page │
│ POST /api/report → receive remote window tasks │
│ GET /api/health → liveness probe │
│ │
│ orphan check (30s) → parent alive? → 3 fails → self-terminate │
└─────────────────────────────────────────────────────────────────────┘
┌─ opencode window #2 (non-owner) ───────────────────────────────────┐
│ bg-tasks process │
│ │
│ serveStdio (MCP stdio) ←─ own LLM tool calls │
│ mini HTTP :9877 │
│ GET /logs/{id} → log page (read-only) │
│ │
│ reporter → every 10s POST /api/report (:9876) push task state │
│ health check → every 5s GET /api/health (:9876) → owner alive? │
│ → takeover :9876 → become new owner → close mini HTTP │
└─────────────────────────────────────────────────────────────────────┘
Owner/Takeover Lifecycle
Window #1 starts → binds :9876 → owner → dashboard online
Window #2 starts → EADDRINUSE → non-owner → mini HTTP :9877 → reporter push
Window #3 starts → EADDRINUSE → non-owner → mini HTTP :9878 → reporter push
Window #1 closes → health check fails → Window #2 jitter 500-1500ms → binds :9876
→ new owner → closes mini HTTP :9877 → tasks become local
Window #2 closes → Window #3 takes over
All windows closed → orphan check triggers → self-terminate
Data Flow
non-owner window owner window
───────────────── ─────────────
bg_start("server", ...) ──┐
bg_start("client", ...) ──┤
│
reporter (10s interval) ◄────┘
│
│ POST /api/report { pid, port, tasks: [...] }
▼
owner dashboard ──► aggregates remote + local tasks
│
├─ GET / → dashboard HTML (window list, running, history)
├─ GET /logs/{id} → local task logs (from MemoryLogStore)
└─ remote task links → http://127.0.0.1:{port}/logs/{id} (mini HTTP)
Logging System
JSONL format written to <install-dir>/.log:
{"ts":"20260809T023839","evt":"owner_ok","pid":30508,"port":9876,"ppid":11972}
{"ts":"20260809T023900","evt":"bg_start","pid":30508,"id":"d26c6305","name":"backend","cmd":"python..."}
{"ts":"20260809T024200","evt":"bg_exit","pid":30508,"id":"d26c6305","exit":0,"ms":60000}
{"ts":"20260809T024503","evt":"orphan_check","pid":30508,"attempt":1,"alive":false}
{"ts":"20260809T030000","evt":"shutdown","pid":30508,"reason":"SIGTERM"}
Auto-rotation: 1MB per file, keeps last 2 backups (.log.0, .log.1).
i18n
Translations are JSON files in lib/i18n/. To add a new language, copy en.json to <lang>.json and translate the values.
lib/i18n/
├── en.json # English (default, fallback)
└── zh.json # 中文
File Structure
bg-tasks-mcp/
├── index.js # entry: MCP stdio + owner detection + orphan check
├── lib/
│ ├── store.js MemoryLogStore (ring buffer, 4MB per key)
│ ├── registry.js TaskRegistry (state machine, onTaskEvent hook)
│ ├── platform.js Windows/Unix adaptation (shell, kill, signal)
│ ├── session.js session ID generation
│ ├── status-http.js HTTP dashboard + routes + multi-window aggregation
│ ├── reporter.js non-owner task state push (10s interval)
│ ├── logger.js JSONL file logger (1MB rotation)
│ ├── i18n.js i18n loader (in-memory cache, en fallback)
│ └── i18n/
│ ├── en.json English translations
│ └── zh.json 中文翻译
├── tools/ 6 MCP tools (bg_start, bg_wait, bg_logs, etc.)
├── resources/ MCP resources (bg://status, bg://{id}/logs, etc.)
└── test/ Node.js native test runner
Environment Variables
| Variable | Description | Default |
|---|---|---|
BG_TASKS_NO_ORPHAN_CHECK |
Set to 1 to disable orphan detection |
off |
Platform
Node.js 18+, Windows/Linux/macOS. Windows uses taskkill instead of POSIX signals.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
E2B
Using MCP to run code via e2b.