</>MCP Agents Market
MCP Server

haft

by m0n0x41d1.4kGoUpdated 2026-08-11

Engineering decisions engine that know when they're stale. Frame, compare, decide — with evidence decay and parity enforcement. For Claude Code, Cursor, Gemini CLI, Codex and more.

Claude CodeCodexCursorGrok CLIGemini CLI

Haft is an MCP server and skill set that adds durable project memory and governance to AI coding agents through the First Principles Framework (FPF). It enables agents to record engineering decisions, compare solution options, track evidence decay, and maintain specifications while enforcing parity rules and human authority boundaries. Haft works with Claude Code, Cursor, Codex CLI, Gemini CLI, Grok, and other MCP-compatible clients, storing a versioned artifact graph that links problems, decisions, specifications, and code together. The tool acts as a governance layer between the human principal, the AI agent, and the codebase, ensuring that binding decisions require explicit operator requests while keeping project reasoning connected to implementation.

Key Features

MCP server with 12 tools for decision recording, problem framing, solution comparison, and evidence tracking
12 auto-triggering and manual skills including h-reason, h-frame, h-explore, h-compare, h-decide, h-verify, and h-status
Evidence decay system with formality levels (F0–F9), congruence levels (CL0–CL3), and automatic staleness detection
Source-native FPF delivery with versioned framework retrieval and project-local governance gates
Code-graph exploration with symbol lookup and concern queries linked to decision records and specifications
Human authority enforcement: binding decisions and work commissions require direct operator requests, not prompt text
Runner-neutral WorkCommission lifecycle for authorized execution tracking without built-in automation
Multi-client support with adapter-specific configurations for Claude Code, Codex, Cursor, Grok, Gemini CLI, Zed, and Hermes

Use Cases

  • 01Recording architectural decisions with evidence, baselines, and validity windows for later verification
  • 02Diagnosing test failures using parallel hypothesis testing with separate Agent subagents to prevent anchoring
  • 03Comparing technology options under explicit parity rules to generate Pareto fronts instead of single winners
  • 04Verifying that past engineering decisions still hold by measuring claims against current baselines and metrics
  • 05Maintaining typed SpecSections for target systems and software systems with drift detection and lifecycle management
  • 06Exploring code context and impact before editing governed files using symbol and concern-based graph queries

Related MCP Servers

View more

haft — FAQ

What is Haft?+

Haft is an MCP server and skill collection that provides project memory and governance for AI coding agents using the First Principles Framework. It records engineering decisions, specifications, and evidence while enforcing human authority boundaries and tracking when information becomes stale.

How do I install Haft?+

Run the installer script with `curl -fsSL https://raw.githubusercontent.com/m0n0x41d/haft/main/install.sh | bash`, then initialize your project with `haft init` for interactive host selection or `haft init --claude`, `--codex`, `--cursor`, etc. for specific clients. Restart your coding agent after initialization.

Which AI clients work with Haft?+

Haft has stable support for Claude Code and Codex CLI/App, with experimental adapters for Grok CLI, Cursor, Gemini CLI, Hermes, Zed, Antigravity, and OpenCode. Each client uses different MCP config files and skill directories as documented in the initialization table.

Do I need API keys or external services?+

No, Haft is local-first and stores its artifact graph in a project-specific SQLite database under `~/.haft/projects/<id>/`. An optional embedding sidecar (haft-embed) may use around 1-2 GB RAM if enabled for semantic recall; you can disable it with `embedding.provider: none` in `~/.haft/config.yaml`.

Is Haft free to use?+

Yes, Haft is open-source and released under the MIT license.

Do I need to learn the entire First Principles Framework to use Haft?+

No, you can install Haft and work normally; agents retrieve only the relevant FPF methods when needed. Use skills like `/h-reason` for deliberate source-supported reasoning or `/h-status` for a compact project overview without learning the full framework upfront.

How do I install haft?+

Open the source repository on GitHub and follow its README. haft is a mcp server — MCP Agents Market links you directly to the official repo.

Is haft free?+

haft is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches