# Seshat

> Structural code intelligence for AI agents. Seshat compiles a codebase into a typed
> symbol graph — every function, route, and table with its real dependency edges, data
> flow, constraints, and git history — served over MCP. Answers are computed from the
> compiled graph, not inferred from likelihood: if Seshat says a function has two
> callers, it has exactly two. Every answer discloses the coverage behind it.

You can verify the core claim right now, without an account: the public demo API is
keyless. See /llms-full.txt for the executable recipe (POST any public GitHub repo,
get a real blast radius back, check it against the source).

## Docs

- [Full agent-readable overview, tool roster, pricing, and verification recipe](https://seshat.papyruslabs.ai/llms-full.txt)
- [Tool roster as JSON](https://seshat.papyruslabs.ai/tools.json)
- [Try it on a public repo (human UI)](https://seshat.papyruslabs.ai/try)
- [Real repos, investigated](https://seshat.papyruslabs.ai/records)

## Install

- MCP server: `npm install @papyruslabsai/seshat-mcp` — 26 tools, works in Claude Code, Cursor, Windsurf, or any MCP harness
- Pricing: first extraction free · first 100 queries free · $0.03/query after

## Company

- [Papyrus Labs](https://papyruslabs.ai) — contact: joseph@papyruslabs.ai