# Verging Labs MCP server

Compare independent test results to see where each memory system works well and where it fails.

This MCP (Model Context Protocol) server gives AI agents direct access to the published [Verging Labs](https://verginglabs.com) benchmarks: the Agentic Memory Index (agent memory tools) and the Agentic Search Index (web search and retrieval tools). Zero dependencies, no telemetry, no account.

## Tools

| Tool | What it does | Cost |
| --- | --- | --- |
| `fetch_memory_index` | Agentic Memory Index rankings: index score with confidence intervals, per-kind pass rates, cost per 1k successful answers, time to ready, retention decay | Free |
| `fetch_search_index` | Agentic Search Index rankings for web search and retrieval tools | Free |

## Use it

Claude Code:

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

Claude Desktop (`claude_desktop_config.json`):

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

Any other MCP client: run `npx -y @verginglabs/mcp` as a stdio server.

No install at all: the two index tools, plus site search and page reading, are served over Streamable HTTP at `https://verginglabs.com/mcp` (manifest at `https://verginglabs.com/.well-known/mcp.json`).

## The data behind it

Rankings come from versioned index releases measured on simulated real work, never self-reported claims. Nothing is blended into one magic number and no provider pays for placement.

- Methodology: https://verginglabs.com/methodology
- Governance: https://verginglabs.com/governance
- Changelog of index releases: https://verginglabs.com/changelog
- API and agent onboarding: https://verginglabs.com/agent-access and https://verginglabs.com/llms.txt

## Free Memory Check

[Request a free assessment](https://verginglabs.com/agent-access) of the memory system you build or use. The team reviews requests; this does not activate an account. For supported maker products, the first test report is free. Users can receive an assessment from published results or request an evaluation of an unmeasured system.

## License

MIT