MCP.so
Sign In
A

Alcove

@epicsagas

About Alcove

Alcove gives AI coding agents on-demand access to your private project documentation — without dumping everything into the context window.

Basic information

Category

Other

Transports

stdio

Publisher

epicsagas

Submitted by

Sangseng Lee

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "alcove": {
      "command": "alcove",
      "args": []
    }
  }
}

Tools

No tools detected

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is Alcove?

Alcove is an MCP server that gives AI coding agents on-demand access to your private project docs. It organizes documentation in a separate directory (one folder per project) and serves it to any MCP-compatible agent over stdio. Docs are never injected into the context window and never leak into public repositories. Built for developers who need agent access to architecture, decisions, runbooks, and PRDs without per-project configuration for every agent.

How to use Alcove?

Install via Homebrew on macOS (brew install epicsagas/alcove/alcove), cargo binstall alcove, cargo install alcove, or clone and make install. Run alcove setup for interactive configuration (docs location, document categories, agent setup). The MCP server starts with the alcove command (called by agents). Use CLI commands like alcove search, alcove validate, alcove doctor for local operations. Agents call tools such as search_project_docs and get_doc_file via stdio MCP.

Key features of Alcove

  • On-demand doc retrieval — agents search, nothing pre-loaded into context
  • BM25 ranked search (tantivy) with automatic grep fallback
  • One doc-repo for multiple projects, organized by project folder
  • Single setup works with any MCP-compatible agent
  • Auto-detects project from current working directory
  • Scoped access — each project only sees its own docs
  • Cross-project search with scope: "global"
  • Standardized doc structure enforced via policy.toml
  • Document validation against team policy

Use cases of Alcove

  • Give agents access to architecture docs without context bloat
  • Search across all projects for patterns like authentication or rate limiting
  • Enforce consistent documentation standards across a team
  • Keep sensitive internal docs off your public GitHub repo
  • Switch between projects and agents without reconfiguring doc access

FAQ from Alcove

How does Alcove differ from using CLAUDE.md?

CLAUDE.md is for short conventions and agent behaviors. Project documentation — architecture, decisions, runbooks, PRDs — doesn't scale in a context file. Alcove is not a replacement; it's the layer CLAUDE.md was never meant to be.

What search algorithm does Alcove use?

Alcove automatically selects BM25 ranked search (powered by tantivy) when the index exists, falling back to grep for exact substring matching. The index builds in the background and updates on file changes.

Does Alcove work with multiple AI agents?

Yes. Alcove works with nine or more agents including Claude Code, Cursor, Claude Desktop, Cline, OpenCode, Codex, Copilot, Antigravity, and Gemini CLI. One setup, all agents share the same access.

Where do my docs live?

Your docs live in a separate directory (DOCS_ROOT) on your machine. They are never stored in your project repositories, keeping sensitive documentation private. Alcove manages them there and serves them to agents over stdio.

How does Alcove detect which project I'm working on?

Alcove detects the project from your terminal's current working directory. You can override this with the MCP_PROJECT_NAME environment variable if your CWD doesn't match a project name in your docs repo.

Comments

More Other MCP servers