╔══════════════════════════════════════════╗
║                                          ║
║   ███╗   ███╗███████╗███╗   ███╗         ║
║   ████╗ ████║██╔════╝████╗ ████║         ║
║   ██╔████╔██║█████╗  ██╔████╔██║         ║
║   ██║╚██╔╝██║██╔══╝  ██║╚██╔╝██║         ║
║   ██║ ╚═╝ ██║██║     ██║ ╚═╝ ██║         ║
║   ╚═╝     ╚═╝╚═╝     ╚═╝     ╚═╝         ║
║                                          ║
║   memory and grounded knowledge          ║
║   for coding agents                      ║
║                                          ║
╚══════════════════════════════════════════╝

Validated project memory (mfm:spec) and grounded tooling discovery (mfm:atlas).

$ why mfm

Chat is a lousy source of truth. Coding agents start every session from zero: architecture buried in scrollback, decisions re-litigated, tools picked from stale training data. The model isn't the bottleneck — the knowledge it works from is.

MFM gives your agent knowledge that survives the session. mfm:spec keeps a validated graph of your system that any agent can read, evolve, and build from. mfm:atlas grounds tooling choices in sourced capability evidence.

MFM never replaces your agent or runs inference on the hot path. You bring the model; MFM supplies deterministic stores, validators, and skills — lines in the sand for the water to follow.

$ products

MFM Spec
project memory
A validated graph of your system — components, features, evaluations, decisions — that any agent can read, evolve, and build from. Local files or hosted MCP, portable either way.
→
Atlas
tooling discovery
Grounded product and tooling discovery: sourced provider capabilities, select and compare, every claim backed by doc evidence instead of stale training data.
→

$ connect

Set up your agent
One MCP server plus installable skills. Copy-paste agent prompt, manual setup for Claude Code, Codex, and Cursor, and the public skill catalog.
→
api.mfm.devGET

The public Atlas read API (/v1/atlas): providers, capabilities, select, and compare — grounded in source evidence. No auth required.

mcp.mfm.devMCP

Model Context Protocol server. Atlas is public today; hosted MFM Spec tools sit behind authenticated project binding.

$ agent briefing

MFM (MadeForMachine) gives coding agents knowledge that survives the session.

Products:
  mfm:spec  A validated spec graph of a system — components, features, evaluations, decisions — to read, evolve, and build from.
  mfm:atlas Grounds product and tooling discovery in sourced capability evidence.

Each service is useful on its own. They are loosely coupled by a shared doctrine: agents work better when context is structured, durable, and actionable. MFM runs no inference; the agent brings the model.

This page is a human-readable visualization. For programmatic access:

  Skills:    https://mfm.dev/skills.json
  Setup:     https://mfm.dev/connect
  REST API:  https://api.mfm.dev/v1/atlas
  MCP:       https://mcp.mfm.dev

Atlas endpoints (public, no auth):
  GET  /v1/atlas/capabilities        — the capability vocabulary
  GET  /v1/atlas/providers           — acquired providers + capability counts
  GET  /v1/atlas/providers/:slug     — one provider's grounded capabilities
  GET  /v1/atlas/providers/:slug/tree — its doc tree (depth-bounded)
  GET  /v1/atlas/select              — match the field by required/excluded capabilities
  GET  /v1/atlas/compare             — pivot a shortlist against a topic, with evidence

Hosted MFM Spec tools are authenticated and project-scoped.
copy this text and paste it to your agent