MOZAIC

Code Intelligence

Your codebase as a queryable graph.

MOZAIC parses your codebase, builds a knowledge graph in SQLite, and exposes it to AI agents via MCP. Functions, dependencies, call chains, and data pipeline lineage — all queryable.

See the interactive demo

mozaic

MOZAIC:

sync_order_batch()

├── validate_order_payload() → libs/validation.py:34

├── transform_order_row() → apps/pipeline/transforms.py:112

└── write_to_warehouse() → apps/pipeline/etl.py:67

└── bq_client.insert() → libs/bq/client.py:201

4 dependencies, 3 modules. No circular references detected.

How it works

1. Parse

AST-level analysis of your codebase. Functions, classes, imports, call chains — extracted from source, not guessed from text. Handles Python, TypeScript, SQL templates, Jinja, and Airflow DAGs.

2. Graph

Builds a SQLite knowledge graph. Nodes are code entities. Edges are relationships — CALLS, IMPORTS, WRITES_TO, READS_FROM. Code-to-code and code-to-data lineage in one graph.

3. Expose

15 MCP tools let any AI agent query the graph. Token-budget-aware responses. The agent asks, MOZAIC answers with source-backed precision.

Language and platform support

MOZAIC starts with the stack where code-to-data lineage matters most. Coverage expands based on design partner needs.

Languages

Pythonsupported

TypeScriptsupported

SQLsupported

Jinja / SQL templatessupported

Goroadmap

Javaroadmap

Rustroadmap

Data Platforms

Airflow DAGssupported

BigQuerysupported

Snowflakeroadmap

dbtroadmap

Sparkroadmap

Connect in under a minute

MOZAIC is an MCP server. If your agent speaks MCP, it speaks MOZAIC. Works with Claude Code, Cursor, Copilot, Windsurf, or any MCP client.

mcp config

{
  "mozaic": {
    "command": "mozaic",
    "args": ["serve", "--project", "."]
  }
}