LabscoMCP early access ↗
olgasafonovaMCP SERVER

Ridge

Point it at a repository and get the architecture back — services, packages, databases, queues and endpoints as a typed graph, plus a drift diff between any two git refs.

by olgasafonova ↗Code Intelligence & Repository IndexingCommunity

LABSCO SUMMARY

Best on the codebase you did not write.

The scan-plus-generate pair gives you a current architecture model of an unfamiliar tree in minutes, which is the step that disappears when an agent writes the change instead of you. Drift is the other half: comparing two refs catches a new database dependency or a bypassed gateway that reads as a small diff in code review. Repeat scans reuse cached analysis for unchanged files, so the second pass is fast.

What it is

An MCP server that reverse-engineers architecture from a directory tree by static analysis. Go is parsed with go/ast, TypeScript and Python with tree-sitter, and markdown vaults through their wiki-links — all producing the same node-and-edge model, with a confidence score on every edge so you can filter heuristic matches out.

What you get

  • 19 tools over one graph model. arch_scan returns the whole architecture; arch_focus narrows to a subdirectory or service
  • Dependency work — arch_dependencies maps internal, external and infrastructure edges, arch_blast_radius returns everything that transitively depends on a target, arch_dataflow traces endpoints through to data stores
  • Shape and explanation — arch_boundaries calls the topology monolith, monorepo or microservices and shows the signals behind the verdict with an ambiguity flag; arch_explain writes it up with code evidence
  • arch_generate renders the graph in 9 formats, including Mermaid, PlantUML, C4, Structurizr, draw.io, Excalidraw and a self-contained D3 force-directed page
  • Drift — arch_drift compares two branches, tags or commits, arch_drift_explain returns a 2-5 sentence narrative you can paste into a PR with no LLM call, arch_diff compares against a saved baseline and arch_snapshot saves one
  • Rules and metrics — arch_validate finds circular dependencies, orphans and layering violations with a baseline mode that grandfathers what already exists; arch_metrics computes coupling, instability and dependency depth; arch_recommend ranks improvements with the metric evidence that triggered each
  • History — arch_history shows how the architecture moved over git history
  • A repo registry so you stop repeating paths: arch_registry_add, arch_registry_list, arch_registry_remove

Requirements

No account and no key — it reads local directories. go install github.com/olgasafonova/ridge/cmd/ridge@latest puts the binary in $GOPATH/bin; you also need a C compiler for the tree-sitter CGo bindings, which is standard on macOS and Linux. Point your client's config at the binary path. Set RIDGE_ALLOWED_DIRS to a colon-separated list of absolute paths if you want scanning restricted to them — without it, ridge will read any directory except a built-in denylist covering /etc, /proc, /sys, /dev and home dotfiles like .ssh and .aws. Scan caches live in ~/.mcp-context/ridge/. MIT.

Setup effort

One command — go install github.com/olgasafonova/ridge/cmd/ridge@latest

See installation options