Claude Code & MCP

Use PromptBeacon from Claude Code, Cursor, or any client that speaks the Model Context Protocol. Ask "does AI recommend my library?" in your editor and get the same numbers as the CLI.

Claude Code plugin

The repository is also a plugin marketplace. In Claude Code:

/plugin marketplace add yotambraun/promptbeacon
/plugin install promptbeacon@promptbeacon

The plugin starts the MCP server with uv (uvx --from 'promptbeacon[mcp]' promptbeacon mcp), so uv must be installed. It adds three skills:

Skill	What it does
ai-visibility	Checks whether AI assistants recommend a product or project and reports score, share of voice and how it was measured
share-card	Creates a README badge, shields.io JSON and light/dark share cards
visibility-ci	Adds a weekly or pull-request GitHub workflow (run it with /promptbeacon:visibility-ci)

Any MCP client

Install the extra and run the server over stdio:

pip install 'promptbeacon[mcp]'
promptbeacon mcp

Register it with Claude Code without the plugin:

claude mcp add promptbeacon -- uvx --from 'promptbeacon[mcp]' promptbeacon mcp

Or in a client's JSON configuration:

{
  "mcpServers": {
    "promptbeacon": {
      "command": "uvx",
      "args": ["--from", "promptbeacon[mcp]", "promptbeacon", "mcp"],
      "env": {
        "OPENAI_API_KEY": "${OPENAI_API_KEY:-}",
        "ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY:-}"
      }
    }
  }
}

Tools

Every tool is a thin wrapper over the Python API and returns compact JSON (no raw model responses).

Tool	Arguments	Returns
providers	none	Which providers and Tavily are configured (true/false, never the key values) and the default mode
scan	brand, category, competitors, providers, prompts, demo, grounded	Score, share of voice and rank, presence, a per-brand table, top sources, models, prompts, cost, top recommendations, measurement_tier
project_scan	project (github:owner/name, pypi:pkg, npm:pkg), category, competitors, providers, prompts, demo, infer_category	The scan result plus a project object with the metadata and the guessed category and competitors, with their basis
sources	brand, category, competitors, providers, demo, grounded (default true)	The domains AI answers cite and whether each cites the brand
share_assets	brand or project, category, competitors, providers, demo, output_dir, png	Paths of badge.svg, badge.json, card.svg, card-dark.svg (and card.png), README snippets, and the summary

A scan without a category returns a warning explaining that the prompts were generic.

Keys and cost

Provider keys reach the server through its environment: the variables in your shell, or the env block of the client configuration (the plugin maps each supported key explicitly). Empty values count as not configured.

• With no key configured, tools default to the keyless demo and the result includes a note that says so. Nothing is billed.
• With keys, scans are real and billed to those keys; results include an estimated cost.
• Passing demo: false without any key is an error rather than a silent fallback.

The server writes nothing to stdout except the protocol. Under the plugin, PromptBeacon's data directory is the plugin's data directory.