# geo-score [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/jianruntech/geo-score  
**GitHub Stars:** 399  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/geo-score

## Description
Free 0-100 score: can AI engines cite a site? Plus citation checks with your own API keys.

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

```json
"mcpServers": {
  "geo-score": {
    "command": "uvx",
    "args": ["geo-score@1.8.2"]
  }
}
```

## Documentation

## What geo-score MCP server does

The geo-score MCP server gives an MCP-compatible agent a structured way to inspect a website’s readiness for citation by AI answer engines. Its first level produces a free score from 0 to 100 using a published, versioned rubric rather than an undisclosed ranking formula. The report also identifies the checks that affect the next scoring tier.

The assessment covers several observable areas, including crawler access, server-rendered content, sitemap and llms.txt signals, structured data, page-type markup, answer passages, headings, freshness, sourced statistics, authorship, knowledge-graph signals, sameAs links, and content shaped for extraction. Some checks are excluded when they require off-site search or human judgement.

## How it works

The server reads public URLs and returns the same three levels available in the command-line interface. `score_site` performs the no-key assessment. A separate question workflow sends one question to selected AI engines through APIs configured with the user’s own keys. The third level watches a question set over time, allowing results to be compared week over week.

The score is separate from citation testing: asking an engine whether it cites a site does not add to the 0–100 result. The rubric distinguishes observable site signals from checks that cannot be measured during a run. Reports include the run duration and explain which checks were full, partial, zero, or not measured.

## Setup and configuration

The geo-score MCP server is a local Python stdio process using the standard library. Python 3.8 or newer is required. The documented client configuration starts version 1.8.2 with `uvx`:

```json
{
  "mcpServers": {
    "geo-score": {
      "command": "uvx",
      "args": ["geo-score@1.8.2", "mcp"]
    }
  }
}
```

This setup requires `uv`, but does not require Git or a build step. The server can also be started from a release wheel or directly from the versioned Git source. No keys are needed for the free scoring level. Provider keys may be supplied through the environment of the process for paid engine queries; the README does not specify their variable names. Add `--read-only` after `mcp` when the server must not spend money.

## Tools and capabilities

The geo-score MCP server exposes these documented capabilities:

- Score one or more public website URLs without an API key.
- Ask one question through configured AI-engine APIs using the caller’s keys.
- Track a question set from one week to the next.
- Return rubric details and the gaps associated with the next score tier.
- Report observations outside the score, such as page-level robots rules, server-HTML links, and hreflang information.

## Limitations and notes

The score measures site signals that can be observed from public URLs; it does not prove that an AI engine will cite a site. Off-site listings, independent mentions, and question coverage may require search or judgement and can remain outside the measured denominator. Chinese-engine readiness is not measured for a non-Chinese-language site.

Engine queries may incur provider charges and require the caller’s API keys. The MCP process is local and uses stdio, so clients must be able to launch a local MCP command. Claude Desktop does not use the shell’s `PATH` for `uvx`; its configuration needs the full executable path.

_Full upstream README: https://allmcps.com/mcp/geo-score/readme_