Skip to content
MCP server

Nacre

Self-hosted knowledge index where each search returns only what the caller may read

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
self-hosted, API, CLI
Runs
Locally
Version
0.41.0
Listing updated
3 days ago

About Nacre

Nacre is an open-source MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.

Nacre's central claim is that its access filter runs inside the index during HNSW traversal, not after ranking. A request for ten results therefore returns ten permitted ones rather than ten minus whatever was stripped. The query builder takes a resolved access plan and has no parameter for a raw filter, and a document you cannot see returns the same response as one that does not exist.

Permissions inherit from workspaces to layers with read, write and admin levels. Revocation takes effect on the next request: a database trigger bumps the organization's permission version and the resolver's cache is keyed on it. Per the README, document-level grants and deny rules are commercial, and the open build refuses them rather than accept a rule it cannot enforce. The core is Apache 2.0, with commercial modules for multi-tenancy, SSO and SCIM, deny rules and SIEM forwarding.

The npm package @nacre.work/mcp runs over stdio and needs a service-account key plus the deployment's database and backend configuration. Retrieval is hybrid, fusing dense vectors and BM25 with an optional cross-encoder rerank, and embeddings can come from any OpenAI-compatible endpoint. It runs under Docker Compose or a Helm chart with no phone-home. Reference connectors cover git, S3, SQL databases, Google Drive, MongoDB and IMAP.

Install

Run in a terminal

claude mcp add nacre --env NACRE_SERVICE_KEY='<NACRE_SERVICE_KEY>' -- npx -y @nacre.work/mcp

Add to claude_desktop_config.json

{
  "mcpServers": {
    "nacre": {
      "command": "npx",
      "args": [
        "-y",
        "@nacre.work/mcp"
      ],
      "env": {
        "NACRE_SERVICE_KEY": "<NACRE_SERVICE_KEY>"
      }
    }
  }
}

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "nacre": {
      "command": "npx",
      "args": [
        "-y",
        "@nacre.work/mcp"
      ],
      "env": {
        "NACRE_SERVICE_KEY": "<NACRE_SERVICE_KEY>"
      }
    }
  }
}

Add to .vscode/mcp.json

{
  "servers": {
    "nacre": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@nacre.work/mcp"
      ],
      "env": {
        "NACRE_SERVICE_KEY": "<NACRE_SERVICE_KEY>"
      }
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "nacre": {
      "command": "npx",
      "args": [
        "-y",
        "@nacre.work/mcp"
      ],
      "env": {
        "NACRE_SERVICE_KEY": "<NACRE_SERVICE_KEY>"
      }
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "nacre": {
      "command": "npx",
      "args": [
        "-y",
        "@nacre.work/mcp"
      ],
      "env": {
        "NACRE_SERVICE_KEY": "<NACRE_SERVICE_KEY>"
      }
    }
  }
}

Configuration

  • NACRE_SERVICE_KEYService-account key for STDIO mode. The server also needs the deployment's full environment (database and backend configuration) - see docs/mcp.md; the key alone is not enough to start it.Required

What it does

Filter inside the index
Access rules apply during traversal so top_k returns k permitted results.
Immediate revocation
A permission version bump makes the very next request see the change.
Inherited permissions
Workspaces and layers pass read, write and admin levels down.
Hybrid retrieval
Dense vectors and BM25 are fused, with optional reranking.
Bring your own embeddings
Any OpenAI-compatible endpoint, bound per layer.
Reference connectors
Six connectors sync git, S3, SQL, Drive, Mongo and IMAP sources.
MCP and REST
Agents use MCP and applications use REST through the same authorization service.

Good for

  • Give agents company search that respects each user's access
  • Prove to an auditor which documents a query could reach
  • Keep a private knowledge index inside your own network
  • Search exact identifiers such as error codes and invoice numbers
  • Sync repositories or buckets into a permissioned index

Pricing

Open source

Apache 2.0 core; commercial modules for multi-tenancy, SSO/SCIM, deny rules and SIEM forwarding (prices not published)

Prices change. Check nacre.work before you buy.

Nacre FAQ

How do I install it?

Clone the repository, copy .env.example to .env and run docker compose with the minimal profile; a Helm chart is also provided.

Is it open source?

The core is Apache 2.0; multi-tenancy, SSO and SCIM, deny rules and SIEM forwarding are commercial modules.

Does it phone home?

No. The README says it stays inside your network with no phone-home.

Does the stdio MCP server work alone?

No. The service key alone is not enough; it needs the deployment's full environment.

Which sources have connectors?

git, S3-compatible stores, Postgres/MySQL/MariaDB, Google Drive, MongoDB and IMAP; there is none for Confluence, SharePoint, Slack, Notion or Jira.

Is Nacre free?

Yes. Nacre is open source, so you can use and self-host it for free.

How do I add Nacre to Claude Code?

Run this command in your terminal: claude mcp add nacre --env NACRE_SERVICE_KEY='<NACRE_SERVICE_KEY>' -- npx -y @nacre.work/mcp

What are the best alternatives to Nacre?

The closest alternatives to Nacre are REA, Codebase Memory, XERJ, Claude Code Ultimate Guide and Nautilus Compass Memory.

Alternatives to Nacre

  1. REA

    MCP server

    CLI and MCP server that lets coding agents reverse engineer apps and binaries

    Runs locally
    • Open source
    • 464K GitHub stars
    • Developer tools
  2. Codebase Memory

    MCP server

    Local MCP server that builds a code knowledge graph for AI coding agents

    Runs locally
    • Open source
    • 48.7K GitHub stars
    • Knowledge and memory
  3. XERJ

    MCP server

    Local search engine for AI agents with cited answers over your code and docs

    Runs locally
    • Open source
    • 19.2K GitHub stars
    • Knowledge and memory
  4. MCP server for searching the Claude Code Ultimate Guide from any MCP client

    Runs locally
    • Open source
    • 6.5K GitHub stars
    • AI models and LLM tooling
  5. Local-first long-term memory layer for AI agents, served over MCP

    Not checked yet
    • 5 tools
    • Remote
    • Open source
    • 1.3K GitHub stars
    • Knowledge and memory
  6. Agent Roast

    MCP server

    Offline review that flags agent changes which weaken tests, assertions or CI

    Runs locally
    • Open source
    • 4 Hacker News points
    • Developer tools