DOCUMENTATION
Your AI agent doesn't know your project. Alcove fixes that.
An MCP server that gives AI agents on-demand access to your private project docs — without dumping everything into the context window, without leaking docs into public repos, and without per-project config for every agent you use.
01 · QUICKSTART
Install via Homebrew (macOS), cargo-binstall, or build from source.
# macOS brew install epicsagas/tap/alcove # Linux / Windows (pre-built binary) cargo binstall alcove # Any platform (build from source) cargo install alcove
Run the interactive setup wizard. It walks you through doc root location, document categories, diagram format, embedding model, and which AI agents to configure.
alcove setup
One command installs the binary and registers the MCP server for Claude Code.
claude plugin install epicsagas/alcove
02 · CORE CONCEPTS
Alcove doesn't inject your docs into context. Agents search for what they need, when they need it. BM25 ranked search returns the most relevant documents first — one result per file, deduplicated and scored.
Private docs organized by project in a single directory. Auto-detects your project from CWD — no per-project config needed. Supports cross-project search with scope: "global".
Combine BM25 full-text search with vector similarity search (powered by fastembed). Download an embedding model during setup, and Alcove automatically uses hybrid ranking for both CLI and MCP queries.
Works with Claude Code, Cursor, Claude Desktop, Cline, OpenCode, Codex, Copilot, Antigravity, and Gemini CLI. Configure once, every agent gets the same access.
03 · REFERENCE
Core CLI commands for managing projects, searching docs, and running the server.
alcove # Start MCP server alcove setup # Interactive setup alcove doctor # Check installation health alcove search "auth flow" # Search current project alcove search "OAuth" --scope global # All projects alcove validate # Validate against policy alcove lint # Broken links, orphans, stale markers alcove promote <file> # Import external note alcove serve # Start HTTP RAG server alcove enable # Register as macOS login item
13 MCP tools available to any connected agent.
search_project_docs # Smart search (BM25 or grep, scoped) get_project_docs_overview # List docs with classification get_doc_file # Read a specific doc list_projects # Show all projects audit_project # Cross-repo audit init_project # Scaffold docs for new project validate_docs # Validate against policy rebuild_index # Rebuild search index check_doc_changes # Detect changed docs lint_project # Semantic lint promote_document # Import from external vault search_vault # Search knowledge base vaults list_vaults # List all vaults
04 · CONFIGURATION
Configuration lives at ~/.config/alcove/config.toml. Set interactively via alcove setup or edit directly.
docs_root = "/Users/you/documents" [core] files = ["PRD.md", "ARCHITECTURE.md", "DECISIONS.md", "CONVENTIONS.md"] [server] host = "127.0.0.1" port = 57384 token = "alcove-a3f7b2..." # auto-generated [memory] reader_ttl_secs = 300 max_cached_readers = 1 model_unload_secs = 600 max_hnsw_cache = 3
Choose from 10 embedding models based on language support, disk size, and quality.
# Set and download a model alcove model set MultilingualE5Small alcove model download # Switch models alcove model set SnowflakeArcticEmbedS alcove rebuild # required after model change
05 · EXAMPLES
When an agent needs project context, it calls search_project_docs. Alcove auto-detects the project from CWD and returns ranked results.
# Agent: "how is auth implemented?" # → Alcove detects project: my-app # → BM25 search: "auth" # ARCHITECTURE.md (score: 0.94) # DECISIONS.md (score: 0.71) # → Agent gets the 2 most relevant docs, not all 12
Search across all projects at once from CLI or via agent MCP tool.
# CLI alcove search "rate limiting patterns" --scope global # MCP (agent) search_project_docs(query="OAuth token refresh", scope="global")