The MAP Framework can be installed in any project to provide powerful AI-driven development capabilities using the Modular Agentic Planner architecture.
- Python 3.11 or higher
- Git (optional, for repository initialization)
- Claude Code CLI
Install the mapify CLI tool globally and use it to set up projects:
# Install mapify CLI
uv tool install --from git+https://github.com/azalio/map-framework.git mapify-cli
# Create a new project with MAP Framework
mapify init my-project
# Or initialize in current directory
mapify init .⚠️ Important: PATH Configuration
After installation, you may need to add UV's bin directory to your PATH.
Check if mapify is accessible:
which mapifyExpected output: /Users/your-username/.local/bin/mapify (macOS/Linux) or C:\Users\your-username\.local\bin\mapify (Windows)
If the command is not found, you need to add ~/.local/bin to your PATH.
UV provides a helper command to automatically configure your shell:
uv tool update-shellThis will update your shell configuration file (.zshrc, .bashrc, etc.) automatically.
If you prefer manual configuration, add the following to your shell configuration file:
For Zsh (macOS default, Linux):
# Add to ~/.zshrc
export PATH="$HOME/.local/bin:$PATH"For Bash (Linux, older macOS):
# Add to ~/.bashrc or ~/.bash_profile
export PATH="$HOME/.local/bin:$PATH"For Fish:
# Add to ~/.config/fish/config.fish
set -gx PATH $HOME/.local/bin $PATHFor Windows (PowerShell):
# Run in PowerShell as Administrator
$userPath = [Environment]::GetEnvironmentVariable("Path", "User")
$newPath = "$env:USERPROFILE\.local\bin"
if ($userPath -notlike "*$newPath*") {
[Environment]::SetEnvironmentVariable("Path", "$userPath;$newPath", "User")
Write-Host "Added $newPath to user PATH"
} else {
Write-Host "$newPath already in PATH"
}After editing your shell configuration file, apply the changes:
# For Zsh
source ~/.zshrc
# For Bash
source ~/.bashrc
# Or simply open a new terminal windowConfirm mapify is now accessible:
mapify --versionExpected output:
mapify-cli version x.x.x
Troubleshooting:
- If
which mapifyshows the path butmapifydoesn't work, check file permissions:ls -la ~/.local/bin/mapify - If using a custom shell or environment, ensure
UV_TOOL_BIN_DIRis not set to a different location - For Docker/CI environments, consider setting
UV_TOOL_BIN_DIR=/usr/local/binfor system-wide access
Run without installing:
# One-time usage
uvx --from git+https://github.com/azalio/map-framework.git mapify init my-projectmapify init my-projectThis will:
- ✅ Create project directory
- ✅ Install MAP agents (including Synthesizer, DebateArbiter, ResearchAgent, FinalVerifier)
- ✅ Add 10 command files plus the skill-backed
/map-learnslash surface - ✅ Configure essential MCP servers
- ✅ Initialize git repository
- ✅ Install branch-scoped
.map/<branch>/workflow runtime used by/map-planand/map-efficient
Note: MAP Framework is designed for Claude Code. All generated agents and commands are optimized for the Claude Code CLI.
Choose which MCP servers to enable:
# All available MCP servers
mapify init my-project --mcp all
# Essential servers only (sequential-thinking, deepwiki)
mapify init my-project --mcp essential
# Specific servers
mapify init my-project --mcp "sequential-thinking,deepwiki"
# No MCP servers
mapify init my-project --mcp none# Initialize in current directory
mapify init .
# Force overwrite existing files in current directory
mapify init . --force# Skip git initialization
mapify init my-project --no-git
# Combine options
mapify init my-project --mcp all --no-gitIf you prefer manual setup:
-
Download the latest release:
wget https://github.com/azalio/map-framework/releases/latest/download/map-kit-template-claude.zip
-
Extract to your project:
unzip map-kit-template-claude.zip -d your-project/ cd your-project -
The structure will be:
your-project/ ├── .claude/ │ ├── agents/ # 11 specialized agents │ │ ├── task-decomposer.md # Decomposes tasks into subtasks │ │ ├── actor.md # Implements code │ │ ├── monitor.md # Validates implementations │ │ ├── predictor.md # Analyzes impact and risks │ │ ├── evaluator.md # Scores solution quality │ │ ├── reflector.md # Extracts lessons │ │ ├── synthesizer.md # Self-MoA: Merges variants │ │ ├── debate-arbiter.md # Opus: Cross-evaluates variants │ │ ├── research-agent.md # Isolated codebase research │ │ ├── final-verifier.md # Adversarial verification (Ralph Loop) │ │ └── documentation-reviewer.md # Reviews technical docs │ ├── commands/ # 10 command-backed slash commands │ │ ├── map-efficient.md # Optimized workflow (recommended) │ │ ├── map-debug.md # Debug workflow │ │ ├── map-fast.md # Minimal workflow (low-risk only) │ │ ├── map-review.md # Review workflow │ │ ├── map-release.md # Release workflow │ │ ├── map-check.md # Quality gates & verification │ │ ├── map-plan.md # Architecture decomposition │ │ ├── map-task.md # Execute one planned subtask │ │ ├── map-tdd.md # Test-first workflow │ │ └── map-resume.md # Resume interrupted workflows │ ├── skills/ │ │ ├── README.md │ │ ├── skill-rules.json │ │ ├── map-planning/ │ │ │ └── SKILL.md # Planning/reference skill │ │ └── map-learn/ │ │ └── SKILL.md # Skill-backed /map-learn workflow │ └── mcp_config.json
Check that everything is installed correctly:
mapify checkOutput should show:
Check Available Tools
● Git version control (available)
● Claude Code CLI (available)
✅ All tools are installed! MAP Framework is ready to use.
After installation, you can use MAP commands in Claude Code:
# Standard production workflow (RECOMMENDED)
/map-efficient Add user authentication with JWT tokens
# Debug an issue
/map-debug Fix API timeout on large file uploads
# Quick low-risk change
/map-fast Implement a small UI tweak
# Review changes
/map-review
# Extract lessons after workflow completion
/map-learnMAP Framework uses slash commands as entry points that coordinate specialized agents in the main Claude Code context:
/map-efficient⭐ - Optimized workflow (5-6 agents): task-decomposer → actor → monitor → predictor (conditional)/map-debug- Diagnostic and fix workflows with agent coordination/map-fast- Minimal workflow (3 agents) — small, low-risk changes (reduced analysis)/map-review- Comprehensive review with Monitor, Predictor, and Evaluator agents/map-check- Quality gates and verification for staged changes/map-plan- Architect phase only: decompose task without implementation/map-release- Package release workflow with validation gates/map-resume- Resume incomplete MAP workflow from checkpoint/map-learn- Extract lessons from completed workflows (implemented as a skill, not a command file)
Note: Agents are invoked automatically by slash commands. Direct agent invocation is not the recommended approach—use the slash commands above for proper workflow orchestration.
If you selected MCP servers during installation, ensure they're configured:
- Complex problem decomposition
- Iterative refinement of solutions
- Edge case discovery
- Read documentation from any GitHub repo
- Analyze architectural patterns
- Learn from production implementations
For enhanced pattern retrieval using semantic similarity instead of keyword matching:
# Install semantic search dependencies
pip install -r requirements-semantic.txtWhat you get:
- 🎯 Meaning-based search (not just keywords)
- 🧠 Synonym understanding: "JWT signature" ≈ "token verification"
- ⚡ Automatic deduplication of similar patterns (90% threshold)
- 💾 Fast embedding cache (
.claude/embeddings_cache/)
Technical Details:
- Model:
all-MiniLM-L6-v2(80MB, 384 dimensions) - Speed: ~3000 sentences/second on CPU
- First run downloads ~500MB model (works offline afterwards)
Fallback: If not installed, MAP uses keyword matching automatically.
Troubleshooting: Common issues include:
- HuggingFace authentication issues (set
HF_TOKENif needed) - Keras 3 compatibility (update to latest sentence-transformers)
- Model download problems (check network connectivity)
To update to the latest version:
# Reinstall mapify with latest version
uv tool upgrade mapify-cli
# Update agents in existing project
mapify init . --forceIf you get zsh: command not found: mapify or bash: mapify: command not found, this is usually a PATH configuration issue.
Diagnosis:
# Check if mapify binary exists
ls ~/.local/bin/mapify
# Check if ~/.local/bin is in your PATH
echo $PATH | grep ".local/bin"Solution 1: Add UV bin directory to PATH (Recommended)
See the PATH Configuration section above for detailed shell-specific instructions, or use UV's automatic setup:
uv tool update-shellThen open a new terminal or run:
source ~/.zshrc # or ~/.bashrc for BashSolution 2: Use full path as workaround
~/.local/bin/mapify --versionSolution 3: Check custom UV_TOOL_BIN_DIR
If you've set a custom UV_TOOL_BIN_DIR, check that location instead:
echo $UV_TOOL_BIN_DIR
ls $UV_TOOL_BIN_DIR/mapifySolution 4: Reinstall mapify
If the binary doesn't exist, reinstall:
# Ensure UV is installed
curl -LsSf https://astral.sh/uv/install.sh | sh
# Reinstall mapify
uv tool install --from git+https://github.com/azalio/map-framework.git mapify-cliVerify the fix:
mapify --version# Check Claude installation
which claude
# If using local installation after migrate-installer
ls ~/.claude/local/claudeCheck that MCP servers are properly configured in your Claude Code settings. The configuration file is at .claude/mcp_config.json.
To remove MAP Framework:
# Remove from project
rm -rf .claude/agents/
rm -rf .claude/commands/
rm .claude/mcp_config.json
rm -rf .claude/embeddings_cache/
# Uninstall mapify CLI
uv tool uninstall mapify-cli- GitHub Issues: https://github.com/azalio/map-framework/issues
- Documentation: https://github.com/azalio/map-framework
- Community: Discussions on GitHub
MIT License - See LICENSE file for details