Skip to main content
Glama
handcraftedbits

jina-reader-mcp

README.md
# jina-reader-mcp

A minimal MCP server [Docker](https://www.docker.com/) image used to access a self-hosted or managed [Jina Reader](https://github.com/jina-ai/reader)
instance.

# Usage

## Starting the Server

```shell
docker run -d --rm \
  --env JINA_API_KEY=<API key> \
  --env JINA_BASE_URL=<URL> \
  --end JINA_TIMEOUT=<timeout>
  -p <port>:8000
  ghcr.io/handcraftedbits/jina-reader-mcp:latest
```

Where:

* `<API key>` is your Jina API key. Omit the `JINA_API_KEY` environment variable if the instance you're using doesn't require one.
* `<URL>` is the base URL of your Jina instance. If the `JINA_BASE_URL` environment variable is omitted, `https://r.jina.ai` will be used.
* `<timeout>` is the default timeout in seconds for all operations. If the `JINA_TIMEOUT` environment variable is omitted, the default of `30` seconds
  will be used.
* `<port>` is the local port to map to. The MCP server will be available at `http://<host>:<port>`.

## Agent Details

Consult your agent harness's documentation for information on how to connect to the MCP server.

The MCP server will expose two tools:

* `web_fetch`: Fetches a single page in either Markdown (default) or JSON formats. Parameters:
  * `content_type` (optional, either `json` or `markdown`): the format of the returned page. By default, this is `markdown`, and the tool description
    encourages agents to use this format.
  * `timeout` (optional): the timeout, in seconds, for the fetch. By default, this is either `30` seconds or the value of the `JINA_TIMEOUT`
    environment variable.
  * `url` (required): The URL of the page to fetch.
  * `with_all_images` (optional): whether or not information about all the images on the page should be returned. By default, this is `false`.
  * `with_all_links` (optional): whether or not information about all the links on the page should be returned. By default, this is `true`.
* `web_fetch_parallel`: fetches up to five (maximum amount recommended by Jina) pages in parallel. Accepts a list of the same parameters as
  `web_fetch`.