Foundry MCP

Deterministic project context management for AI coding assistants

An MCP (Model Context Protocol) server that enables AI assistants like Claude and Cursor to maintain persistent project context, specifications, and task lists across development sessions. Stores all context in ~/.foundry/ to keep your codebase clean.

Installation

Install from crates.io

cargo install foundry-mcp

Setup with AI Assistants

After installation, configure Foundry with your AI development environment:

# For Cursor
foundry install cursor

# For Claude Code
foundry install claude-code

# Verify installation
foundry status

That's it! Foundry MCP tools are now available in your AI assistant.

Installation includes helpful templates:

• Cursor: Gets .cursor/rules/foundry.mdc with edit_commands guidance and workflow examples
• Claude Code: Gets ~/.claude/agents/foundry-mcp-agent.md with edit_commands usage and intelligent defaults

Why Foundry?

• 🎯 Persistent Context: Never lose project context between AI sessions
• 🗂️ Clean Separation: Project specs stored outside your codebase in ~/.foundry/
• 📋 Structured Planning: Organized specifications with task breakdowns
• 🔄 Resume Work: Pick up complex development exactly where you left off
• 🛠️ Deterministic: Reliable file operations instead of error-prone AI prompting

Problem Statement

Current AI coding assistant workflows suffer from critical issues:

• Inconsistent file management: Prompt-driven systems create files in unpredictable locations
• Project directory pollution: Context management files clutter the actual codebase
• Context loss: No reliable way to pause/resume complex development tasks across sessions
• Error-prone prompting: Relying on natural language instructions for file system operations leads to inconsistent behavior

Solution Overview

Foundry MCP provides a set of MCP tools that enable deterministic project and specification management through a centralized file system outside of project directories.

Core Value Propositions

1. Deterministic operations: MCP tools eliminate prompt-driven file system errors
2. Clean project separation: Context files stored outside project directories
3. Persistent context: Natural pause/resume functionality through structured file storage
4. Hierarchical organization: Project-level context with individual specs and task lists
5. Simplified installation: One-command setup with PATH-based reliability and automatic configuration

MCP Tools

Foundry provides 9 MCP tools that enable comprehensive project management for AI assistants:

Project Management

• create_project: Create new project with vision, tech stack, and summary
• analyze_project: Create project structure by analyzing existing codebases
• load_project: Load complete project context for LLM sessions
• list_projects: List all available projects with metadata

Specification Management

• create_spec: Create timestamped specification with task breakdown
• load_spec: Load specification content with project context
• update_spec: Edit spec files using intent-based edit commands with precise anchors and idempotent updates
• delete_spec: Delete existing specification and all its files

Content & Workflow

• validate_content: Validate content against schema requirements
• get_foundry_help: Get comprehensive workflow guidance and examples

How It Works

Foundry stores structured project context in ~/.foundry/:

~/.foundry/my-project/
├── vision.md          # Product vision and goals
├── tech-stack.md      # Technology decisions
├── summary.md         # Quick context summary
└── specs/
    └── 20250826_143052_user_auth/
        ├── spec.md        # Feature requirements
        ├── task-list.md   # Implementation checklist
        └── notes.md       # Design decisions

Benefits: Clean codebase separation • Persistent across git operations • Chronological feature tracking

AI Assistant Benefits

When you work with AI assistants like Claude or Cursor, Foundry provides:

🔄 Session Continuity

Day 1: "Build a task management app"
→ AI creates project with vision, tech stack, feature specs

Day 5: "Let's work on authentication"
→ AI loads complete context, knows exactly what you're building

📋 Structured Development

• Project-level context: Vision, tech decisions, and architecture choices persist
• Feature specifications: Detailed requirements with implementation task lists
• Progress tracking: Task lists update as AI completes work

🧠 Enhanced AI Performance

• No repeated explanations: AI loads full project context in seconds
• Consistent decisions: Technology choices and architecture preserved
• Better code quality: Specifications guide implementation details
• Reduced hallucination: Structured context prevents AI from making assumptions

✏️ Deterministic Edit Commands

• Intent-based commands: set_task_status, upsert_task, append_to_section only
• Precise selectors: task_text (exact checkbox text) and section (case-insensitive headers)
• Idempotent updates: Safe to re-run commands without duplication or side effects
• Smart error recovery: Candidate selector suggestions with exact match requirements

🤝 Collaborative User Experience

• Option-based guidance: All tools provide "You can..." suggestions instead of directive commands
• Content creation acknowledgment: Tools explicitly recognize AI assistant's role in content generation
• Workflow efficiency: Smart guidance for optimal tool selection and usage patterns
• User decision-making control: Preserves user agency while providing helpful guidance

MCP Tools Available

Once installed, AI assistants have access to these tools:

• create_project - Create new project with vision, tech stack, and summary
• analyze_project - Create project from existing codebase analysis
• load_project - Load complete project context for AI sessions
• list_projects - List all available projects with metadata
• create_spec - Create timestamped specification with task breakdown
• load_spec - Load specification content with project context
• update_spec - Edit spec files using intent-based commands: set_task_status, upsert_task, append_to_section
• delete_spec - Delete existing specification and all its files
• validate_content - Validate content against schema requirements
• get_foundry_help - Get workflow guidance and examples

Development

Building

git clone https://github.com/cafreeman/foundry-mcp.git
cd foundry-mcp
cargo build --release

Testing

cargo test

Running the MCP Server

# Start MCP server
cargo run -- serve

# With verbose logging for debugging
cargo run -- serve --verbose

CLI Testing (Optional)

Test MCP tools from the command line:

# Basic workflow test
cargo run -- mcp create-project test-app --vision "Test app" --tech-stack "Rust" --summary "Testing Foundry"
cargo run -- mcp list-projects
cargo run -- mcp load-project test-app

Contributing

1. Fork the repository
2. Create a feature branch
3. Add tests for new functionality
4. Ensure cargo test passes
5. Submit a pull request