i
mako_help
When you know the task but not the tool sequence. Returns the ordered recipe, suggested args, and verification path.
Wire Mako into your agent.
agentmako is a local-first MCP server for AI coding
agents — Claude Code, Codex CLI, Cursor, Cline. Each setup guide is a
five-minute walk: install, attach a project, paste an MCP config,
restart the agent, run your first mako_help or
context_packet.
Quickstart
A two-step prerequisite, then any agent.
// install & attach a real project
$ npm install -g agentmako $ cd /path/to/your/project $ agentmako connect . --no-db $ agentmako --version $ agentmako doctor # expect green $ agentmako tool list # confirm tools are registered
Use --no-db for the first run. You can add Postgres /
Supabase awareness later with interactive
agentmako connect . or
--db-env DATABASE_URL --yes for CI.
Pick your agent
Four harnesses. Same MCP server.
Claude Code
Plain MCP wiring, or the bundled mako-ai plugin with
discovery / trace / graph / database skills.
Codex CLI
Drop into ~/.codex/config.toml. Codex picks Mako tools
up on next run. Works with codex exec and the TUI.
Cursor
Per-project .cursor/mcp.json or global config. Composer
and Agent mode both see Mako tools.
Cline
↗ VS Code extension. Settings → MCP Servers, paste the snippet, reload. Works inside any workspace you've attached.
First commands to try
Once it's wired, start with these.
ii
context_packet
When you have a coding task. Returns the file, routes, symbols, findings, risks, modes, and what to read next.
iii
file_preflight
When a file is about to change. Returns findings, stale diagnostic flags, conventions, recent runs, and ack history in one call.
Tip
Copy the agent guidance from
AGENTS.md
into your project's CLAUDE.md / AGENTS.md. It
tells the agent when to call which Mako tool — without it
they default to grep.
Concepts
How agentmako thinks about a codebase.
What is MCP?
↗ The Model Context Protocol — what it is, where it came from, why one server runs against every coding agent.
Context packets
↗
What context_packet returns: ranked files, related
routes, active findings, and a "read next" pointer.
Reef Engine
↗ Durable facts and findings that persist across agent sessions — so the next turn does not re-discover everything.
Freshness model
↗ How Mako labels evidence as live, fresh_indexed, stale, contradicted, or unknown — and why "stable" is not "fresh."
Recipes
Walk-throughs for the common agent tasks.
Trace an auth flow
↗ Bug report says "manager role check is failing." Walk a route to its handler, find the RLS, fix the bug.
Refactor safely
↗ Renaming an exported function across a real codebase without shipping broken callers.
Schema-aware edits
↗ Add a Postgres column with RLS, find every code site that touches the table, audit tenant isolation.
Debug with context
↗
A real "hydration error" debug session — start from
context_packet, narrow with freshness, fix.
Reference
Tools, CLI, dashboard, FAQ.
Tool catalog
↗ Every typed MCP tool, by category. Read or mutation, with input and output examples.
CLI reference
↗
Every agentmako command — connect, doctor, mcp,
project, providers, and the optional harness suite.
Dashboard
↗ Optional Vite + React dashboard. Browse projects, run tools, inspect Reef state, chat with the harness — local-only.
FAQ
↗ MCP fundamentals, codebase context, agentmako specifics, and troubleshooting.
Troubleshooting
↗ "MCP server fails to start", "tools don't appear", "stale results" — diagnose and fix in one place.
What's next
Read the thinking, watch the traffic.
The blog covers the cost story, what grep can't do for an agent, multi-tenant audit patterns, and cross-session memory with Reef. Practical posts; no hype.