BasketWatch MCP Server

A Model Context Protocol server that exposes the BasketWatch Irish grocery data API as tools your AI agent can call directly. Works with Claude Desktop, Continue.dev, Cursor, and any other MCP-compatible client.

~50,000 Irish supermarket SKUs across Aldi, Tesco, SuperValu and Dunnes Stores. Refreshed every night. Now queryable by Claude.

Try it with no key at all

Install it, point it at the API, ask a question. With no key you get a live 50-row sample on every tool, straight from this morning's scrape. Real prices, real promotions, enough to see whether the data is what you need before paying anything.

Paying for it: credit packs

Access is metered per record returned, at EUR 0.001 per record, bought up front as credits. No subscription, no monthly minimum, and credits never expire.

Buy one at <https://basketwatchireland.com/pricing> and you are emailed a key. Put it in BASKETWATCH_API_KEY and this server starts using it.

Your credits work here exactly as they do anywhere else. There is no separate MCP plan, no surcharge and no second balance to keep track of. One record returned costs one credit whether it came from curl, your own code, or your assistant calling a tool. Every response reports what it cost and what is left, and GET /api/credits shows the balance and ledger for free.

Calls that return no rows are free. A question that finds nothing costs nothing.

Agents can buy their own

An autonomous agent can purchase a credit pack itself, with no human, no browser and no card entry, by paying over MPP with a Stripe shared payment token. It reads the price at /api/credits/mpp, receives a signed payment challenge, pays, and gets back a funded key it can use immediately. See the API reference.

Bespoke access

Bulk historical exports, scheduled delivery, custom matching or an SLA are a conversation rather than a signup: [email protected].

The MCP server itself is free and open source. The data behind it is what you pay for.

What your agent can do

Once installed, your agent has 8 grocery-aware tools:

Example questions your agent can answer:

• "Where can I buy Heinz Baked Beans cheapest this week?"
• "Build me a €40 weekly grocery list across the 4 supermarkets."
• "Which products got cheaper at Tesco this week?"
• "What promotions are running on chocolate at Dunnes right now?"
• "Track the price of Brennans bread for the next 12 weeks."

Install

pip install basketwatch-mcp

Or from source:

git clone https://github.com/basketwatch/basketwatch-mcp.git
cd basketwatch-mcp
pip install -e .

Configuration

Important: by default the MCP server imposes no client-side rate limits — paid subscribers get whatever throughput their key allows on the BasketWatch origin. The two BASKETWATCH_MCP_*_LIMIT env vars are escape valves for use cases like:

• Giving a Claude Desktop install to someone who shouldn't burn through

the family / team API quota

• Self-imposed budget caps during evaluation

The origin's per-key rate limit (enforced server-side) is the authoritative throttle. Trial keys get tight limits; paid-subscriber keys get high or unlimited throughput.

When a client-side limit is configured AND hit, every tool returns:

{
  "error": "Client-side rate limit reached: ...",
  "limit_hit": true,
  "hint": "Email [email protected] to get an API key with higher / unlimited usage."
}

Hook it up to Claude Desktop

Edit your Claude Desktop config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add a basketwatch entry under mcpServers:

{
  "mcpServers": {
    "basketwatch": {
      "command": "basketwatch-mcp",
      "env": {
        "BASKETWATCH_API_BASE": "https://basketwatch.fly.dev",
        "BASKETWATCH_API_KEY":  "your-api-key-here"
      }
    }
  }
}

Restart Claude Desktop. You'll see "basketwatch" listed as an available MCP server in Claude's MCP indicator. Ask Claude "What's the cheapest 1L of milk in Ireland?" — it'll use the tools automatically.

Hook it up to Continue.dev / Cursor

These clients use the same MCP protocol over stdio. Add a server block to your ~/.continue/config.json (Continue) or equivalent Cursor config:

{
  "mcpServers": {
    "basketwatch": {
      "command": "basketwatch-mcp",
      "env": {
        "BASKETWATCH_API_BASE": "https://basketwatch.fly.dev",
        "BASKETWATCH_API_KEY":  "your-api-key-here"
      }
    }
  }
}

Local development

git clone https://github.com/basketwatch/basketwatch-mcp.git
cd basketwatch-mcp
python -m venv .venv && source .venv/bin/activate  # or .venv\Scripts\activate on Windows
pip install -e .

export BASKETWATCH_API_BASE="https://basketwatch.fly.dev"
export BASKETWATCH_API_KEY="your-key"

# Run the server directly to test (it speaks MCP over stdio — type JSON-RPC
# requests at it or use `mcp dev` from the MCP SDK for interactive testing).
python basketwatch_mcp.py

Data freshness

• All four retailers are refreshed every night.
• Each tool response includes a scrape_date field per row so the agent

knows exactly when each price was captured.

• Higher-cadence pulls (daily / twice-weekly) are available with a direct

subscription — email [email protected].

Not affiliated

BasketWatch is not affiliated with Aldi Ireland, Tesco Ireland, Musgrave SuperValu or Dunnes Stores. Data is collected from publicly available sources for lawful market research and price comparison purposes.

License

MIT.