Vibe Coding Discoverlean-ctx
View on GitHubLeanCTX — Context Intelligence for AI systems.
Local-first context intelligence layer and MCP server for AI coding agents: it compresses file reads and shell output, routes and remembers context across sessions, and tracks token savings, cost and accepted outcomes via a live dashboard.
Use Cases
Cut input token spend for coding agents like Claude Code, Cursor, Copilot, Codex and Gemini CLICompress file reads and shell output (git, npm, cargo, docker, kubectl, terraform) before sending to the modelPersist session memory and task facts across chats and context compactionServe context tools to agents over MCP (ctx_expand, ctx_refactor, ctx_proof, ctx_agent)Track token savings, budgets, CPAO and cost via a local dashboard and savings ledgerBenchmark compression gains with Shadow Mode baseline comparisonsHybrid BM25 + embedding + graph-proximity code search over a repositoryLSP-powered rename, references and go-to-definition for refactoringBuild PR context packs and multi-agent handoff bundlesReversible compression with content-addressed recovery of pruned payloadsFetch and compress web pages, PDFs and YouTube transcripts into citation-backed context
Built With
- Language
- Rust
- Frameworks
- MCP · tree-sitter · LSP · rust-analyzer · gopls · pylsp · typescript-language-server
Tags
context-engineering · token-optimization · mcp-server · ai-coding · coding-agents · context-compression · prompt-cache · session-memory · knowledge-graph · rust · local-first · lsp · tree-sitter · cost-tracking · developer-tools · context-layer