Skip to content

Sankshep

Sankshep

Sankshep (संक्षेप) is Sanskrit for "the concise essence" — a summary that keeps the substance and drops the excess.

Maximum context, minimum tokens — with the benchmarks to prove it.

Sankshep is a proprietary MCP server, written in C#/.NET 10, that gives MCP clients — VS Code Copilot Chat, Claude Code, Claude Desktop, Cursor — token-minimized, relevance-ranked codebase context, plus a persistent, per-repo memory of facts and decisions.

Everything runs locally: embeddings via ONNX Runtime, vector search via sqlite-vec, no code or content leaves your machine and there is no telemetry by default.

Why it exists

An AI assistant is only as good as the context it's given — but whole files burn tokens and bury the answer. Sankshep sends the model the small, relevant slice instead: it retrieves the chunks that matter and AST-minimizes them (strips comments, collapses non-target method bodies), then reports how many tokens it saved.

The full surface — ten MCP primitives (eight tools, one prompt and one resource), each documented with real captured I/O in the tool reference:

  • Minimize — ranked, comment-stripped, body-collapsed context under a token budget (get_context).
  • Search — semantic (nearest-neighbor) search over a local embedding index you build with index_repo (search_code).
  • Map — namespaces, public types, and method signatures for a directory (summarize_repo).
  • Remember — a per-repo, branch-scoped fact store that survives across sessions and clients (remember / recall).
  • Decisions — export decision-tagged memory to DECISIONS.md (export_decisions).
  • Measure — cumulative token compression, per tool, no dollar guesswork (token_report).
  • Compose — a grounded one-shot prompt from minimized code + memory (the prompt composer).

Get started

  • Quickstart — install → connect → index → first grounded answer → see the savings, in ~5 minutes.
  • Install — dotnet tool install -g sankshep, client configs, HTTP, and environment variables.
  • Tool reference — every tool's arguments and a real captured request/response.
  • How it works — the three-layer model, MCP, and where Sankshep fits.
  • Deployment — Docker, Windows Service, systemd, Kubernetes, air-gap.
  • Troubleshooting · Security & privacy · Architecture — when something misbehaves, what stays local, and the engineering behind it.

Proprietary, free to use

Sankshep is proprietary software, free to install and use (including commercially). Its source is private and it does not accept external pull requests — but the tool itself stays a public, free-to-install package. See Feedback to reach the maintainer.