Skip to content

Nikita-Guzenko/pilot-mcp

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🛩️ pilot-mcp

Project management autopilot for AI coding agents.

Give your AI agent (MiniMax, Claude Code, Cline, Cursor, Windsurf) a structured workflow: task tracking, progress persistence, build validation, and code quality enforcement — all through MCP tools.

The Problem

AI coding agents lose context between sessions. They forget what they built, what's next, and what rules to follow. You end up re-explaining the same things every session.

The Solution

pilot-mcp gives your agent a persistent project management layer:

  • 📋 Task tracking — Agent reads tasks, completes them in order, never skips
  • 💾 Progress persistence — Survives context resets. Agent picks up exactly where it left off
  • 🔨 Build validation — One tool call to compile and get parsed errors
  • 🔍 Code quality rules — Define forbidden patterns, agent checks automatically
  • 📊 Status dashboard — Progress bar, file count, current task at a glance

Quick Start

# Initialize in your project
cd my-project
npx pilot-mcp init

# This creates:
# pilot/progress.md  — Live progress tracker
# pilot/tasks.md     — Your task list (edit this!)
# pilot/rules.md     — Code quality rules (edit this!)

Then add to your agent's MCP config:

{
  "mcpServers": {
    "pilot": {
      "command": "npx",
      "args": ["pilot-mcp", "serve", "/path/to/your/project"]
    }
  }
}

Tools

Tool What It Does
get_current_task Reads progress.md — where the project stands right now
get_task_details(n) Full instructions for task N from tasks.md
complete_task(n, files, notes) Marks task done, updates progress, sets next task
add_issue(text) Logs a bug or issue in progress.md
build_project() Compiles the project, returns parsed errors or success
validate_file(path) Checks one file against rules.md patterns
validate_all_files() Scans ALL code files against rules.md
project_status() Dashboard: progress bar, file count, current task
read_pilot_file(name) Reads any file from pilot/ directory

How It Works

You (human)                    Agent (AI)                     pilot-mcp
    │                              │                              │
    │  "Start building"            │                              │
    │─────────────────────────────>│                              │
    │                              │  get_current_task()          │
    │                              │─────────────────────────────>│
    │                              │  ← progress.md contents      │
    │                              │  get_task_details(1)         │
    │                              │─────────────────────────────>│
    │                              │  ← task instructions         │
    │                              │                              │
    │                              │  ... does the work ...       │
    │                              │                              │
    │                              │  build_project()             │
    │                              │─────────────────────────────>│
    │                              │  ← ✅ BUILD SUCCEEDED        │
    │                              │  validate_all_files()        │
    │                              │─────────────────────────────>│
    │                              │  ← ✅ 0 violations           │
    │                              │  complete_task(1, [...], "") │
    │                              │─────────────────────────────>│
    │                              │  ← progress.md updated       │
    │                              │                              │
    │                              │  get_task_details(2)         │
    │                              │─────────────────────────────>│
    │                              │  ... continues autonomously  │

Agent Configuration

MiniMax Mini-Agent

Add to ~/.mini-agent/config/mcp.json:

{
  "mcpServers": {
    "pilot": {
      "command": "npx",
      "args": ["pilot-mcp", "serve", "/path/to/your/project"]
    }
  }
}

Claude Code

Add to .claude/settings.json:

{
  "mcpServers": {
    "pilot": {
      "command": "npx",
      "args": ["pilot-mcp", "serve", "/path/to/your/project"]
    }
  }
}

Cline (VS Code)

Add to Cline MCP settings:

{
  "mcpServers": {
    "pilot": {
      "command": "npx",
      "args": ["pilot-mcp", "serve", "/path/to/your/project"]
    }
  }
}

Writing Tasks

Edit pilot/tasks.md:

## Task 1 — Project Setup

### Goal
Create the project scaffold with all config files.

### Steps
1. Initialize the project
2. Install dependencies
3. Create folder structure

### Validation
- Project compiles without errors
- All config files present

### Files
src/index.ts, package.json, tsconfig.json

Writing Rules

Edit pilot/rules.md and add patterns inside the rules block:

console\.log\( ||| Remove console.log before committing
any\b ||| Avoid 'any' type — use proper TypeScript types
TODO ||| Resolve all TODOs before marking task complete
\.font\(\.system\( ||| Use named fonts from design system
Color\.blue\b ||| Use design system colors instead of Color.blue

Rules are JavaScript regex patterns. The ||| separator splits pattern from error message.

Build Auto-Detection

build_project() automatically detects:

Project Type Detection Build Command
Xcode .xcodeproj xcodebuild
Xcode Workspace .xcworkspace xcodebuild -workspace
Node.js package.json npm run build or tsc
Rust Cargo.toml cargo build
Go go.mod go build ./...
Make Makefile make
CMake CMakeLists.txt cmake --build
Python requirements.txt py_compile

Or pass custom_command to override.

First Prompt for Your Agent

Copy-paste this to start your agent:

You have MCP tools from "pilot" available.

Your workflow for EVERY session:
1. Call get_current_task — see where we left off
2. Call get_task_details for the current task number
3. Do the work described in the task
4. Call build_project to verify it compiles
5. Call validate_all_files to check code quality
6. If errors: fix them, repeat steps 4-5
7. Call complete_task with task number, files created, and notes
8. Move to next task — repeat from step 2
9. Do not stop until context runs out or all tasks are done

Start now.

License

MIT — Built by 731Labs
Follow us on X: @NikGuzenko

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors