Pathfinder MCP Server
Enforces a three-phase gated workflow (Research → Plan → Implement) for complex coding tasks with explicit human approval at each phase transition and automatic context management to maintain focus.
README
Pathfinder MCP Server
A FastMCP server implementing a three-phase gated workflow: Research → Plan → Implement (RPI).
Overview
Pathfinder enforces intentional context compaction and explicit human approval at each phase transition. It helps maintain focus during complex coding tasks by:
- Research Phase: Gather information, explore codebase, document findings
- Plan Phase: Create structured implementation plan with explicit phases
- Implement Phase: Execute plan phases with progress tracking
Installation
For End Users (Once Published)
# No installation needed - use uvx to run directly
uvx pathfinder-mcp
For Local Development
# Clone the repository
git clone <repository-url>
cd pathfinder-mcp
# Install dependencies
uv sync
# Install as a global uv tool (editable mode)
uv tool install --editable .
# Install dev dependencies (optional)
uv sync --dev
What does uv tool install --editable do?
- Installs the command globally (available system-wide)
- Links to your source code (changes take effect immediately)
- Isolated from other projects
- No need for complex paths in MCP configs
Configuration
Environment Variables
| Variable | Default | Description |
|---|---|---|
PATHFINDER_SESSIONS_DIR |
~/.pathfinder-sessions |
Session storage path |
PATHFINDER_TRANSPORT |
stdio |
Transport: stdio or sse |
PATHFINDER_HOST |
127.0.0.1 |
SSE server host |
PATHFINDER_PORT |
8080 |
SSE server port |
PATHFINDER_MAX_TOKENS |
128000 |
Max context tokens |
Cursor MCP Config
For Published Package (Recommended):
{
"mcpServers": {
"pathfinder": {
"command": "uvx",
"args": ["pathfinder-mcp"]
}
}
}
For Local Development (After uv tool install --editable .):
{
"mcpServers": {
"pathfinder": {
"command": "pathfinder-mcp",
"env": {
"PATHFINDER_SESSIONS_DIR": "/Users/yourusername/.pathfinder-sessions"
}
}
}
}
SSE Transport (Remote Server):
{
"mcpServers": {
"pathfinder": {
"command": "pathfinder-mcp",
"env": {
"PATHFINDER_TRANSPORT": "sse",
"PATHFINDER_HOST": "0.0.0.0",
"PATHFINDER_PORT": "8080"
}
}
}
}
Note: Use absolute paths in environment variables (e.g.,
/Users/username/...instead of~). The~expansion may not work reliably in all MCP clients.
Tools
| Tool | Phase | Description |
|---|---|---|
health_check |
Any | Server health and context status |
start_research |
- | Initialize research phase |
save_research |
Research | Save research findings |
start_plan |
Research → Plan | Transition to planning (requires approval) |
save_plan |
Plan | Save implementation plan |
implement_phase |
Plan/Implement | Execute implementation phase |
compact_context |
Any | Compress session context |
Resources
| URI | Description |
|---|---|
pathfinder://templates/research |
Research template |
pathfinder://templates/plan |
Plan template (Cursor format) |
pathfinder://templates/checklist |
Implementation checklist |
pathfinder://templates/progress |
Progress tracking template |
Prompts
| Prompt | Description |
|---|---|
generate_plan_from_research |
Create plan from research |
compact_session |
Generate compressed summary |
resume_session |
Restore session context |
Workflow
1. start_research("Build auth system")
└─> Creates session, research.md
2. save_research(findings) [repeat as needed]
└─> Appends to research.md
3. start_plan() [asks for confirmation]
└─> Creates plan.md template, transitions to PLAN
4. save_plan(plan_content)
└─> Validates & saves plan
5. implement_phase() [asks for confirmation]
└─> Executes phase, updates progress.md
6. compact_context() [when utilization >60%]
└─> Creates session_summary.md
Session Artifacts
~/.pathfinder-sessions/
{session_id}/
research.md # Research findings
plan.md # Implementation plan
progress.md # Implementation progress
session_state.json # Session snapshot
session_summary.md # Compacted summary
Architecture
src/pathfinder_mcp/
├── server.py # FastMCP server + tools
├── config.py # Environment configuration
├── state.py # Phase state machine
├── session.py # Session persistence
├── artifacts.py # Markdown file writer
├── context.py # Token monitoring
├── logger.py # Structured logging
├── errors.py # Error handling
├── handlers/ # Modular phase handlers
│ ├── base.py # BaseHandler ABC
│ ├── research.py # ResearchHandler
│ ├── plan.py # PlanHandler
│ └── implement.py # ImplementHandler
└── tools/ # Tool implementations
├── research.py
├── plan.py
├── implement.py
└── compact.py
Development
Use just to run common development tasks:
# See all available recipes
just --list
# Run tests
just test
# Run tests with coverage report
just test-coverage
# Lint & format code
just check
# Fix linting issues
just lint-fix
# Run the server locally
just run
# Build the package
just build
Manual Commands
If you prefer to run commands directly:
# Run tests
uv run pytest
# Run tests with coverage
uv run pytest --cov=pathfinder_mcp
# Lint & format
uv run ruff check src/ tests/
uv run ruff format src/ tests/
# Run server locally
uv run pathfinder-mcp
Pre-commit Hooks
Setup automatic linting and formatting on every commit:
# Install pre-commit hooks (one-time)
just setup-hooks
# Or manually:
uv run pre-commit install
# Run pre-commit on all files
just pre-commit-all
# Pre-commit will automatically run on `git commit`
Context Management
Pathfinder tracks token usage and provides warnings:
- 60% utilization: Suggests compaction
- 70% utilization: Warns strongly
- compact_context: Resets by creating summary
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。