Scan once.
Query forever.
Orstrum walks your TypeScript repository and builds a structured impact graph — every file, every export, every import edge. Serve it through MCP. Your agent gets a database query instead of a file-system expedition.
Three commands. Complete codebase intelligence.
orstrum scan Build the graph
Walks your repository with a full AST parser and tsconfig awareness. Resolves path aliases and barrel re-exports — building the complete dependency graph in a local SQLite database. No network required.
orstrum push Sync to the cloud
Uploads the graph to your cloud workspace. Only structure leaves your machine: file paths, export signatures, import edges, and optional purpose summaries. Raw source is never transferred.
orstrum serve Query on demand
Starts the MCP server against your local graph or cloud workspace. Any MCP-compatible agent — Claude Code or your own — queries eight tools. "Who imports this file?" is one call, not fifty sequential greps.
Structure and meaning,
not source.
Orstrum extracts four entity types: repositories, files (nodes), exports, and import edges. Each edge captures the resolved dependency — not a text approximation. Barrel re-exports and TypeScript path aliases are traced through the AST, not grepped.
Each file can carry a purpose summary: an LLM-generated description of what the file does and why it exists. Summaries are computed once and stored, not regenerated on every query. A compromised cloud instance exposes an architectural map, not your codebase.
Read the schema →I'll check the dependency graph.
Found 3 files that import auth.ts:
orstrum:get_importers same call · same result
Eight tools. One interface,
local or cloud.
The MCP server exposes eight tools for structured graph retrieval and blast-radius analysis. All tools accept source paths — absolute, repo-relative, or shorthand — so an agent never has to manage internal IDs.
-
orstrum_get_summaryLLM purpose + top callers and imports for a file -
orstrum_get_exportsExported symbols with kind and signature, from the AST -
orstrum_get_importsOutbound imports with resolved cross-file edges -
orstrum_get_callersEvery indexed file that imports a given file -
orstrum_searchKeyword search across stored purpose summaries -
orstrum_list_filesAll indexed files, optionally filtered to one repo -
orstrum_analyze_impactBlast-radius analysis: walks the caller graph and reasons over affected files -
orstrum_get_guidanceRouting guidance for agents — when to use Orstrum vs. grep
Capabilities runtime agents cannot replicate.
Structural accuracy
A full AST parser with tsconfig resolution. Barrel files and path aliases are resolved to real edges — not text approximations from grep.
Persistent graph
The graph survives across sessions. A file's importer count last sprint versus this sprint is data that does not exist on disk. Only a persistent observer has it.
Team-shared index
Push once; the whole team queries the same graph. One agent's structural discovery does not expire when its conversation ends.
Zero raw source
Only structure leaves your machine: paths, export signatures, import edges, and purpose summaries. The cloud graph is an architectural map, not a codebase copy.
Cross-repo edges
edges.package_name is already stored. When two
repositories share a package, that dependency crosses repo boundaries.
Cross-repo traversal is on the roadmap.
Free local CLI
The CLI scanner and local MCP server are free to use. Your code never leaves your machine — only the graph syncs to the cloud, never raw source. Cloud sync is the paid layer.
The local graph is always free.
- Local graph (SQLite)
- Unlimited repositories
- All eight MCP tools via
orstrum serve - Community support
billed $240 / seat / year
- Cloud sync via
orstrum push - Hosted MCP endpoint
- Team graph sharing
- Purpose summaries in cloud
- Priority support
The best time to understand
a codebase is before you need to.
Orstrum makes that understanding permanent.