# Sibyl — Deep Research MCP

> Sibyl — Deep Research MCP is a free MCP server for web search and scraping. Local research workspace that tracks sources, claims and gaps for your AI agent. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Free
- **Platforms:** CLI, Linux, macOS, Windows
- **Website:** https://yichenwu.dev/projects/sibyl
- **Repository:** https://github.com/chriswu727/sibyl
- **MCP Registry name:** io.github.chriswu727/sibyl
- **Categories:** web search and scraping, knowledge and memory
- **Listing updated:** 2026-11-15

## About

Sibyl does not search or reason on its own. Your agent does the thinking and searching, and Sibyl keeps a persistent research workspace beside it: a plan, an evidence store of fetched passages, and records of conclusions marked supported, disputed or unresolved. Its README says it calls no other generative model and needs no second model or search API key.

A distinctive check rejects fact claims whose quotes do not appear verbatim in the cited passage. Passages carry stable identities and hashes, repeated copies of the same text can be identified, and counterevidence and open gaps stay in the record. Reports export as Markdown, JSON or PDF, and research can be resumed from a local task. The current source exposes eight tools, including start_research, continue_research, assess_research, complete_research and save_report.

The project is candid about its state. Development is paused as of 28 October 2026, no stable release is planned, and two evaluations showed source support of 79.4% against 74.9% for built-in search, at about twice the cost and within noise. Published packages contain an older architecture than the source checkout.

## Features

- **Persistent research plan:** Keeps research questions across calls and sessions.
- **Quote verification:** Rejects claims whose quotes are not verbatim in the cited passage.
- **Evidence tracing:** Gives passages stable identities, hashes and provenance.
- **Counterevidence tracking:** Records disputed or unresolved conclusions and gaps.
- **Report export:** Writes Markdown, JSON or PDF reports.

## Good for

- Build a cited comparison that includes counterevidence
- Continue an investigation across several sessions
- Hand findings to a colleague with their evidence trail
- Check that a report's quotes match their sources

## Install

### Claude Code (Terminal)

```sh
claude mcp add sibyl -- uvx sibyl-research
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "sibyl": {
      "command": "uvx",
      "args": [
        "sibyl-research"
      ]
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "sibyl": {
      "command": "uvx",
      "args": [
        "sibyl-research"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "sibyl": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "sibyl-research"
      ]
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "sibyl": {
      "command": "uvx",
      "args": [
        "sibyl-research"
      ]
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "sibyl": {
      "command": "uvx",
      "args": [
        "sibyl-research"
      ]
    }
  }
}
```

## Tools

- `start_research`: Starts a research task
- `continue_research`: Adds a page URL to the evidence store
- `assess_research`: Checks claims against exact quotes
- `complete_research`: Finishes a research task
- `save_report`: Saves a cited report

## FAQ

### Does it need another model or API key?

No. The default workflow uses your own assistant and needs no extra model or search key.

### Is it still developed?

No. Development is paused and no stable release is planned.

### Does it beat built-in agent search?

Two evaluations did not show a clear benefit; the difference was within run-to-run noise.

### What does it need to run?

Python 3.11 or newer and a host that supports local stdio MCP.

### Is Sibyl — Deep Research MCP free?

Yes. Sibyl — Deep Research MCP is free to use.

### How do I add Sibyl — Deep Research MCP to Claude Code?

Run this command in your terminal: claude mcp add sibyl -- uvx sibyl-research

### What are the best alternatives to Sibyl — Deep Research MCP?

The closest alternatives to Sibyl — Deep Research MCP are Scrapling MCP Server, Codebase Memory, XERJ, Firecrawl MCP Server and Claude Code Ultimate Guide.

## Alternatives

- [Scrapling MCP Server](https://ai-powered.com/mcp/scrapling-mcp-server): Scrapling MCP Server is an open-source MCP server for web search and scraping.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Firecrawl MCP Server](https://ai-powered.com/mcp/firecrawl-mcp-server): Firecrawl MCP Server is a freemium MCP server for web search and scraping.
- [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.
- [SearXNG Search](https://ai-powered.com/mcp/searxng-search): SearXNG Search is an open-source MCP server for web search and scraping.
- [Nautilus Compass Memory](https://ai-powered.com/mcp/nautilus-compass-memory): Nautilus Compass Memory is an open-source MCP server for knowledge and memory.
- [note1](https://ai-powered.com/mcp/note1): note1 is an MCP server for productivity.

---
Source: https://ai-powered.com/mcp/sibyl-deep-research-mcp
