Astria MCP server

astria

Knowledge graph builder for codebases

Docs | Getting started | CLI Reference | Architecture | Worked examples | Changelog | Release notes

Understand a codebase before you touch it. astria turns any folder into a queryable knowledge graph — deterministic AST extraction in Rust, optional local-embedding semantics, zero API keys, everything on your machine.

astria is inspired by the Python Graphify project's core idea — turn a corpus into a queryable knowledge graph — but it is an independent, from-scratch implementation: a deterministic, offline-first Rust/tree-sitter pipeline, not a fork or a port. astria is not affiliated with, sponsored by, or endorsed by the Graphify project or Graphify Labs.

Find source-grounded code context before changing a repository. Astria retrieves symbols, file locations, and relationships from the latest committed graph snapshot through the same CLI and MCP query engine. Updates reuse cached AST extraction and reconcile references across the current corpus; inferred name matches remain distinct from declared facts.

Retrieval quality and delivered context cost are measured separately. The paired benchmark methodology records file recall, source-grounded symbol checks, exact response tokens and failures. The latest results include improvements and regressions. Historical full-corpus/query token ratios are size diagnostics, not measured savings over targeted source search. A measured first-shot comparison against a deterministic question-derived rg-plus-reads baseline is checked in at worked/external-baseline/: ~207 delivered tokens per question put the defining file first for 8/8 external questions, while the baseline's 4,000-token responses never ranked it first — quality per delivered token against a single-pass search floor, not a claim about expert iterative search.

Code definitions retain their scoped identities during deduplication. Extraction includes assigned JS/TS functions and Python implementations behind overload declarations. Query text uses exact o200k_base budgets shared by CLI and MCP. After upgrading, run astria update . to refresh extraction and restore previously merged definitions; start fresh pagination because cursors now count both node and edge records.

Three things a folder full of files can't give you:

Shortened. The full README is on GitHub.