How Pharaoh Works

Pharaoh parses your codebase using tree-sitter — a deterministic parser with zero hallucination risk — into a Neo4j knowledge graph. Functions, files, modules, imports, exports, call chains, HTTP endpoints, cron handlers, database access patterns, and environment variables are all extracted as nodes and edges. Your AI tool connects via Model Context Protocol (MCP) and queries the graph for structural context before writing or modifying code.

Parsing takes approximately 60 seconds for a 50K LOC TypeScript project. The graph re-indexes automatically via GitHub webhook on every push to the default branch.

Analysis Tools

Pharaoh exposes 13 analysis tools via MCP. 5 are always free; the rest require Pro for full results (free tier receives real preview data).

Codebase Map

Codebase Map — full structural overview with module dependencies, hot files, and endpoint mapping

Module Context

Module Context — complete module profile with functions, DB tables, endpoints, and external callers

Search Functions

Search Functions — find existing code across the entire codebase before creating duplicates

Blast Radius

Blast Radius — trace all downstream callers up to 5 hops before refactoring

Query Dependencies

Query Dependencies — trace forward, reverse, and circular dependencies between modules

Check Reachability

Check Reachability — verify exports are wired to production entry points

Vision Docs

Vision Docs — cross-reference PRDs and CLAUDE.md specs against implementation

Vision Gaps

Vision Gaps — find specs without code and complex code without specs

Cross-Repo Audit

Cross-Repo Audit — compare two repositories for structural duplication and drift

Consolidation Opportunities

Consolidation Opportunities — detect duplicate logic, parallel consumers, and signature twins

Unused Code

Unused Code — graph-based dead code detection with text-reference backup verification

Test Coverage

Test Coverage — per-module coverage with untested high-complexity function flagging

Regression Risk

Regression Risk — score every function by complexity, exposure, churn, and caller count

Architecture

Deterministic tree-sitter parsing into Neo4j graph. No LLM in the analysis pipeline — zero hallucination risk, reproducible results, no per-query API costs. Graph queries trace transitive dependencies up to 5 hops deep, resolving relationships invisible to file-level grep or import scanning.

PR Guard

Automated structural checks on every pull request. Catches unreachable exports, assesses blast radius of changes, scores regression risk, verifies test coverage, and detects introduced duplication. Runs as a GitHub Check Run — advisory on Free, blocking on Pro.

FAQ