Chronicle MCP — Knowledge Graph

Self-learning code analysis tool for coding agents (Claude Code, Codex CLI, ...). Builds a knowledge graph of your codebase — data models, services, endpoints, dependencies — and answers questions like "what breaks if I change X?"

Install

npm install -g @alexdx/chronicle-mcp

Setup with Claude Code

claude mcp add chronicle -- chronicle mcp serve

Or add to your Claude Code MCP config (~/.claude.json):

{
  "mcpServers": {
    "chronicle": {
      "command": "chronicle",
      "args": ["mcp", "serve"]
    }
  }
}

Setup with Codex CLI

chronicle setup codex

This registers the server in ~/.codex/config.toml (plus a SessionStart hook that re-primes the Chronicle entry point after /clear or compaction), writes usage guidance to ~/.codex/AGENTS.md (and to the project's AGENTS.md when run inside a project), and installs /chronicle-scan, /chronicle-impact, ... custom prompts. Equivalent manual config:

[mcp_servers.chronicle]
command = "chronicle"
args = ["mcp", "serve"]
startup_timeout_sec = 30   # journal sync on open can exceed Codex's 10s default
tool_timeout_sec = 180     # resolve/commit on large graphs can exceed the 60s default

Chronicle detects the connected client: Claude Code gets the parallel (subagent) scan workflow; clients without subagents (Codex, Cursor) automatically get a single-agent workflow. Query tools work identically everywhere.

Usage

Open Claude Code in any project and say:

chronicle scan          — Full project scan
chronicle data          — Analyze data models
chronicle language      — Domain language glossary
chronicle impact X      — What breaks if I change X?
chronicle deps X        — What depends on X?
chronicle services      — Service architecture
chronicle status        — Dashboard URL + graph stats

What it does

• Scans your codebase (NestJS, Prisma, GraphQL, Kafka, etc.)
• Builds a multi-layered knowledge graph (data → code → contracts → services)
• Tracks dependencies, impact paths, domain language
• Self-learns: discovers patterns, reports gaps, improves with each scan
• Admin dashboard with visual graph at http://localhost:<auto-port>

The graph is git-native: every mutation is recorded in an append-only event journal (.depbot/events/<domain>.jsonl) that you can track in git, and the SQLite db is a derived cache rebuilt from it anytime (chronicle journal rebuild). Two teammates' scans merge through a normal git merge — after a pull, the graph syncs automatically on the next command. See the repository README for the full journal workflow.

Links

• Repository