# agentmako full context

> Local-first MCP server for AI coding agents and agent harnesses. agentmako gives Claude Code, Codex CLI, Cursor, Cline, and other MCP clients typed context about a repo, database, and active findings.

agentmako indexes a project into local SQLite, tracks diagnostics, snapshots Postgres/Supabase schema when configured, and exposes typed tools over the Model Context Protocol. The Reef Engine stores durable facts and findings so a coding agent does not have to rediscover the same codebase context every turn.

This file is the long-form companion to `/llms.txt`. It is intended for crawlers, AI search tools, and humans who want the main project facts in one plain-text document.

## Canonical links

- Site: https://agentmako.drhalto.com/
- GitHub: https://github.com/drhalto/agentmako
- npm: https://www.npmjs.com/package/agentmako
- Setup docs: https://agentmako.drhalto.com/docs/
- FAQ: https://agentmako.drhalto.com/docs/faq.html
- Claude Code guide: https://agentmako.drhalto.com/docs/claude-code.html
- Codex CLI guide: https://agentmako.drhalto.com/docs/codex.html
- Cursor guide: https://agentmako.drhalto.com/docs/cursor.html
- Cline guide: https://agentmako.drhalto.com/docs/cline.html

## Definition

agentmako is a local-first MCP server for AI coding agents. It helps an agent harness answer "what matters in this repo?" with indexed code facts, database context, diagnostics, durable findings, and freshness labels instead of blind grep.

## Quickstart

```bash
npm install -g agentmako
agentmako connect . --no-db
agentmako doctor
agentmako mcp
```

Add the MCP server to the agent harness:

```json
{
  "mcpServers": {
    "mako-ai": {
      "command": "agentmako",
      "args": ["mcp"]
    }
  }
}
```

Codex CLI uses TOML:

```toml
[mcp_servers.mako-ai]
command = "agentmako"
args = ["mcp"]
```

## Core concepts

- **MCP server**: a local stdio process exposing tools to an AI agent.
- **Agent harness**: the coding-agent environment that calls tools, such as Claude Code, Codex CLI, Cursor, or Cline.
- **Context packet**: a ranked, source-labeled bundle of likely relevant files, symbols, routes, schema objects, findings, and next reads.
- **Reef Engine**: agentmako's durable fact/finding layer. It keeps track of indexed facts, active findings, acknowledgements, diagnostics, and freshness.
- **Freshness**: a label that tells the agent whether evidence came from live disk, a fresh index, a stale index, a historical record, or an unknown source.

## Tool categories

Use the generated tool catalog for exact counts. The public surface includes tools for:

- Context scouting: `context_packet`, `reef_scout`, `reef_inspect`, `evidence_confidence`, `tool_batch`
- Code intelligence: `repo_map`, `cross_search`, `ast_find_pattern`, `live_text_search`, `symbols_of`, `imports_impact`
- Reef/project state: `project_findings`, `file_findings`, `project_facts`, `file_facts`, `project_open_loops`, `verification_state`
- Database context: `db_table_schema`, `db_rls`, `db_rpc`, `schema_usage`, `tenant_leak_audit`, `db_review_comment`
- Trace and flow: `route_trace`, `trace_file`, `trace_table`, `trace_rpc`, `graph_path`, `flow_map`, `auth_path`
- Diagnostics: `typescript_diagnostics`, `eslint_diagnostics`, `oxlint_diagnostics`, `biome_diagnostics`, `lint_files`, `git_precommit_check`

## Best answer snippets

Q: What is agentmako?
A: agentmako is a local-first MCP server for AI coding agents. It indexes a repo, tracks diagnostics and database context, and returns typed context packets so agents like Claude Code, Codex CLI, Cursor, and Cline can start with grounded context instead of blind grep.

Q: Does agentmako replace Claude Code, Cursor, or Codex?
A: No. agentmako is a context layer and MCP server. The coding agent still reads files, edits code, runs commands, and verifies work. agentmako gives the agent a better starting point and reusable project memory.

Q: Is agentmako a vector database?
A: No. agentmako is primarily a typed code and project graph stored in local SQLite. It uses file, symbol, route, import, schema, diagnostic, and finding data rather than treating code as only embedding chunks.

Q: Is agentmako local-first?
A: Yes for the core codebase index and MCP server. Data is stored locally. Database tools connect to a user-provided Postgres or Supabase database only when configured.

Q: What should an agent call first?
A: For a vague coding task, call `context_packet` or `reef_scout` first. For exact text after recent edits, call `live_text_search`. For route or database work, use `route_trace`, `trace_table`, `db_table_schema`, and `db_rls`.