Recipes
Build a Childcare Agent
Wire CribScore MCP to your agent runtime and answer real childcare questions with cited evidence in under thirty minutes.
Goal#
Build an agent that can: search providers by ZIP, compare two candidates, read jurisdiction trust before quoting metrics, and cite an inspection PDF when asked for evidence. Stack: remote MCP for the fastest start.
Prereqs#
Have an MCP-capable agent client (Claude Desktop, Cursor, your own runtime). Have a CribScore API key. That's it.
- API key — generate from dashboard.
- Agent client — anything that speaks MCP streamable HTTP.
- Remote MCP URL — https://api.cribscore.co/mcp
Wire the MCP config#
Drop the streamable-HTTP block into your client config. The agent now has access to every CribScore tool.
{
"cribscore": {
"type": "streamable-http",
"url": "https://api.cribscore.co/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}Programmatic agent (Python / TypeScript)#
If you are building a custom agent runtime rather than using a client like Claude Desktop, use an MCP SDK to call tools directly. The pattern is the same: list tools, call by name, pass the same arguments the REST API accepts.
# pip install httpx mcp-client
from mcp_client import MCPClient
client = MCPClient(
transport="streamable-http",
url="https://api.cribscore.co/mcp",
headers={"Authorization": "Bearer YOUR_API_KEY"},
)
tools = client.list_tools()
result = client.call(
"cribscore_search_facilities",
arguments={"zip_code": "90003", "limit": 5},
)
for facility in result["data"]:
print(facility["name"], facility.get("safety_score"))import { createMcpClient } from "@modelcontextprotocol/sdk/client";
const client = await createMcpClient({
transport: "streamable-http",
url: "https://api.cribscore.co/mcp",
headers: { Authorization: "Bearer YOUR_API_KEY" },
});
const tools = await client.listTools();
const result = await client.callTool({
name: "cribscore_search_facilities",
arguments: { zip_code: "90003", limit: 5 },
});
console.log(result.data.map((f: { name: string }) => f.name));Question patterns the agent should handle#
A useful childcare agent answers four shapes of question. Test each one before shipping; if any one fails, fix the prompt — not the API.
- Search — 'list licensed centers near 90003 with safety score above 80'
- Compare — 'compare safety and price for these two facilities'
- Trust — 'is California decision-grade right now?'
- Evidence — 'what was the most recent inspection finding for facility X?'