# AsterMind MCP

> AsterMind MCP is a free MCP server for ai models and llm tooling. Local MCP server that reranks and trims RAG context to cut tokens before an LLM call. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Free
- **Platforms:** CLI
- **Website:** https://astermindai.com
- **Repository:** https://github.com/astermindai/astermind-mcp
- **MCP Registry name:** io.github.AsterMindAI/astermind-mcp
- **Categories:** AI models and LLM tooling, developer tools
- **Listing updated:** 2026-11-17

## About

AsterMind MCP works on the retrieval side of a RAG pipeline. You hand it candidate passages and a query, and it scores them locally so you only forward the few that matter. The README says it runs on the AsterMind Community Edition ELM engine with no network calls, no API keys and no data leaving the machine.

The project publishes its own numbers: on a six-scenario benchmark that cut 10 candidates down to the top 3, average context-token saving was 66.9% and a passage containing the answer survived in every scenario. It is equally frank about the trade-off: strict recall of all relevant passages was 55.6%, so multi-hop questions should keep five or six chunks.

The README lists known limits openly. Reranking is lexical (TF-IDF), so pure synonym gaps can be missed; classification is weak out of the box; embeddings are character-level and suit near-duplicate detection rather than deep semantic similarity. It runs on Node 18+ with no GPU.

## Features

- **Context reranking:** rerank_documents scores and orders candidate passages against a query.
- **Context filtering:** filter_context keeps the top-K or above-threshold passages.
- **Token-budget compression:** compress_context reranks and trims a block to fit a token budget.
- **Token accounting:** count_tokens and estimate_savings give BPE counts and before/after deltas.
- **Confidence signals:** Every tool returns a confidence flag and says when the engine is weak.

## Good for

- Cut the retrieved context sent to an LLM in a support-KB RAG pipeline
- Estimate token savings before choosing how many chunks to keep
- Detect near-duplicate passages in a corpus
- Fit retrieved text into a fixed token budget before prompting

## Install

### Claude Code (Terminal)

```sh
claude mcp add astermind-mcp -- npx -y @astermind/astermind-mcp
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "astermind-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@astermind/astermind-mcp"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "astermind-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@astermind/astermind-mcp"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "astermind-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@astermind/astermind-mcp"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "astermind-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@astermind/astermind-mcp"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "astermind-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@astermind/astermind-mcp"
      ]
    }
  }
}
```

## Tools

- `rerank_documents`: Score and order candidates against a query
- `filter_context`: Keep the top-K or above-threshold passages
- `compress_context`: Rerank and trim a context block to a token budget
- `count_tokens`: Exact BPE token count for any text
- `estimate_savings`: Before/after token delta for a filtering choice
- `semantic_search`: Rank a corpus against a query
- `detect_language`: Identify the language of text
- `classify_text`: Label text into supplied categories
- `generate_embeddings`: Character-level vector for text
- `compare_texts`: Similarity between two texts

## FAQ

### Does any data leave my machine?

The README says there are no network calls and no API keys; processing is local.

### How much context does it save?

On the project's six-scenario benchmark, keeping the top 3 of 10 passages saved 66.9% of tokens on average.

### What are the main limitations?

The reranker is lexical, classification is weak, and embeddings are character-level.

### How do I run it?

With npx @astermind/astermind-mcp on Node 18+, or by adding the same command to an MCP client config.

### Is AsterMind MCP free?

Yes. AsterMind MCP is free to use.

### How do I add AsterMind MCP to Claude Code?

Run this command in your terminal: claude mcp add astermind-mcp -- npx -y @astermind/astermind-mcp

### What are the best alternatives to AsterMind MCP?

The closest alternatives to AsterMind MCP are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Ouroboros](https://ai-powered.com/mcp/ouroboros): Ouroboros is an open-source MCP server for developer tools.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [Concord MCP](https://ai-powered.com/mcp/concord-mcp): Concord MCP is a freemium MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/astermind-mcp
