# donsetch [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/dondai44423/donsetch  
**GitHub Stars:** 1062  
**npm Downloads (last month):** 11403  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/donsetch

## Description
MCP server that gives AI agents the web: fetch, search, crawl and screenshot from one local Rust binary. Zero API keys, zero accounts, Chrome-true TLS.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "donsetch": {
    "command": "npx",
    "args": ["-y","donsetch"]
  }
}
```

## Documentation

## What donsetch does

donsetch exposes four web-research capabilities to MCP clients: fetching pages, searching the web, crawling linked content, and taking screenshots. It is distributed as a local Rust binary rather than a hosted service, so an agent can perform research through one process and the same functionality can also be used from the command line.

The server is designed for search and data extraction workflows where agents need to locate sources, retrieve page content, follow a site or document structure, and inspect rendered pages visually. Its tool surface is intentionally small, with the four tools together using about 2,369 tokens in the listed `tools/list` JSON representation.

## How it works

The fetch system implements its own HTTP and extraction components, including HTTP/2 handling, connection pooling, cookie persistence, TLS session resumption, conditional revalidation, and TCP Fast Open. Its TLS behavior uses Chrome's native BoringSSL engine, including Chrome-originated ClientHello characteristics, rather than attempting to imitate Chrome with a static fingerprint.

For pages protected by a challenge, a browser can solve the challenge and return cookies to the faster fetch tier. The browser generally does not retrieve the page content itself. Fetch behavior also adapts over time: cookie lifetimes are adjusted, repeatedly blocking sites can enter cooldown, and known walls can be pre-solved during search. Health and improvement information is available through `doctor` and status output.

Search runs multiple providers in parallel, combines results through cross-engine agreement, and applies local semantic reranking. It supports a keyless mode and optional bring-your-own-key providers. Search results and links can be represented by compact handles, while options such as `focus`, `toc`, `section`, `must_contain`, and `since_last` help limit returned content.

## Setup and configuration

Install the platform-specific binary through npm:

```bash
npm install -g donsetch
```

The package downloads a prebuilt release binary and verifies it with SHA256, so a compiler or other build tool is not required. macOS and Linux users can instead install through Homebrew:

```bash
brew tap dondai44423/donsetch && brew install donsetch
```

After installation, run `donsetch doctor` for grouped health checks. `doctor --deep` adds live egress, provider, and captive-portal probes; `doctor --fix` attempts mechanical repairs; `doctor --json` adds machine-readable results; and `doctor --mcp` detects clients and prints registration blocks using absolute paths.

## Tools and capabilities

- Fetch web pages and extract content through the local fetch tier.
- Search across more than 10 providers with consensus and semantic reranking.
- Crawl sites or linked content for multi-page research tasks.
- Capture screenshots for rendered-page inspection.
- Process PDFs by combining extracted glyphs with rendered pixels and automatically applying OCR to scanned documents.
- Reduce agent output with focused extraction controls and reusable search or link handles.

## Limitations and notes

donsetch is distributed as a local executable and is intended to run on the user's machine. Its keyless search mode does not require API keys or accounts, although bring-your-own-key provider configurations are supported. The project is written in Rust and provides a standalone CLI in addition to MCP connectivity.

The released product described by the project is the V4 line; the V5 redesign is being developed privately, while V4 updates continue. Client integrations named by the project include Claude Code and Cursor, alongside other MCP clients and standalone integrations such as Pi and DeepSeek Harness. The project is licensed under AGPL-3.0.

_Full upstream README: https://allmcps.com/mcp/donsetch/readme_

