# Open Agent Search [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/jayanth-mkv/open-agent-search  
**GitHub Stars:** 2  
**npm Downloads (last month):** 622  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/open-agent-search

## Description
Search the web, news, images, videos and books, and extract page content for AI agents.

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

```json
"mcpServers": {
  "open-agent-search": {
    "command": "npx",
    "args": ["--yes","open-agent-search@1.0.5"]
  }
}
```

## Documentation

## What the Open Agent Search MCP server does

The Open Agent Search MCP server provides a shared retrieval layer for agents and applications. Its MCP surface supports text, image, video, news, book, and unified searches. Agents can also fetch the content of one page or multiple pages, making the server useful for workflows that search first and then inspect selected sources.

Search is handled through the `ddgs` library, so the service does not require an account with Open Agent Search or a search-provider API key. It can run as a local process for one coding agent or as an HTTP service for applications, shared environments, REST clients, and remote MCP connections.

## How it works

The Open Agent Search MCP server offers two connection paths. A local stdio process can be started with `uvx` and the pinned package version, allowing an MCP client to launch and stop the process as needed. The same Python implementation can be run as a persistent HTTP service.

The HTTP deployment listens on port 8000 by default and provides REST routes under `/api`, interactive documentation at `/docs`, and streamable HTTP MCP at `/ai/mcp`. REST search routes include separate endpoints for text, images, videos, news, books, and unified results. Content routes handle single-page and multi-page fetching.

Because queries still travel to upstream providers through DDGS, local hosting does not make searches anonymous. Deployment logs, network rules, and provider terms should be reviewed when handling sensitive queries.

## Setup and configuration

The Open Agent Search MCP server requires Node.js 22+ and `uv` when using the npm launcher. The launcher uses a pinned Python core, while `uv` manages Python 3.12 and the required dependencies. The documented npm commands start either the MCP process or the HTTP service.

A Python-only setup can use `uvx --python 3.12 --from open-agent-search==1.0.5 oas-mcp` for stdio, or install the package as a persistent `uv` tool and run `open-agent-search` for HTTP. `uv` can download Python 3.12 if it is not already installed. The first run downloads dependencies.

For source-based deployment, the repository documents `uv sync` followed by `uv run open-agent-search`. Docker Compose and Vercel deployment paths are also provided. No environment variables or API keys are specified as required by the provided setup instructions.

## Tools and capabilities

The Open Agent Search MCP server exposes these MCP-capable operations:

- Search text, images, videos, news, books, or all supported categories together.
- Fetch content from one page.
- Fetch content from multiple pages.
- Use the same search and content functions through local stdio MCP or HTTP-based MCP.
- Access the HTTP API with REST clients when MCP is not the consuming interface.

## Limitations and notes

The server depends on upstream search providers accessed through DDGS, so results and availability depend on that retrieval path. Self-hosting controls where the service runs but does not conceal queries from upstream providers. The documentation recommends choosing stdio for an agent that only needs local tools and HTTP for shared or remote workloads. MCP tool parameters and response shapes are documented separately in the project’s API references.

_Full upstream README: https://allmcps.com/mcp/open-agent-search/readme_