MCP Server

Unihra ships a built-in MCP (Model Context Protocol) server that lets AI assistants — Cursor, Claude Desktop, or any MCP-compatible client — run SEO analyses and query results without any manual setup.

Installation

pip install "unihra[mcp]"   # requires Python 3.10+

Cursor / Claude Desktop Config

Add to your MCP configuration file:

Claude Desktop

{
  "mcpServers": {
    "unihra": {
      "command": "python",
      "args": ["-m", "unihra.mcp_server"],
      "env": {
        "UNIHRA_API_KEY": "YOUR_KEY_HERE"
      }
    }
  }
}

Cursor

{
  "mcpServers": {
    "unihra": {
      "command": "python",
      "args": ["-m", "unihra.mcp_server", "--key", "YOUR_KEY_HERE"]
    }
  }
}

Available Tools

Tool	Description
unihra_health	Check API availability
unihra_get_limits	Show API key balance and daily limits
unihra_analyze	Run SEO analysis (standard or extended)
unihra_list_results	List saved analysis results
unihra_delete_result	Delete a saved result
unihra_get_page_status	List requested URLs included in the analysis and not included
unihra_get_gaps	Get Umbrella Analysis gaps with optional filtering
unihra_get_anchors	Get anchor analysis
unihra_get_word_actions	Get TF-IDF word actions (add/increase/decrease)
unihra_get_ngrams	Get phrase patterns
unihra_get_triplets	Get Knowledge Graph entities and gaps (extended mode)
unihra_get_page_structure	Get heading structure and meta tags

How It Works

1. You tell the AI: "Compare my page X with competitors Y and Z"
2. The AI calls unihra_analyze — analysis runs server-side (30–120 seconds)
3. A compact summary is returned immediately with a result_id
4. The AI calls specific unihra_get_* tools to retrieve focused data slices

Results are saved locally in ~/.unihra_mcp/results/ and persist between sessions.

Cost Awareness

The AI is instructed to use standard mode (1 credit) by default and escalate to extended mode (5 credits) only when you explicitly ask for Knowledge Graph analysis or fact-level content briefs. Use unihra_get_limits to check your balance.

Manual Start

UNIHRA_API_KEY=your_key python -m unihra.mcp_server

# With options:
python -m unihra.mcp_server \
  --key YOUR_KEY \
  --retries 3 \
  --results-dir /tmp/unihra_results

Example Prompts

"Compare https://mysite.com/product with https://comp1.com and https://comp2.com"
"What words should I add to my page to rank for 'buy mattress online'?"
"Do a fact coverage audit — what entities do competitors cover that I miss?"
"Show me the heading structure of all analyzed pages"