Platform / MCP
Give your agent the whole practice market.
MCP is a way for an agent to call tools. The PracticeMCP server exposes the practice record as tools, and each field can come back with the quote the agent should cite instead of inventing.
Config
One server entry.
The same shape works in Claude Desktop, Claude Code, Cursor, and other clients that read an MCP config. Replace the key. Do not commit it.
{
"mcpServers": {
"practicemcp": {
"type": "http",
"url": "https://mcp.practicemcp.com/mcp",
"headers": {
"Authorization": "Bearer pb_live_replace_me"
}
}
}
} Tools
| Tool | What the agent can do |
|---|---|
| search_practices | Filter the market by state, city, ZIP, specialty, or text. Returns a page of locations and a cursor. Costs credits per record. |
| get_practice | One location by id, with its clinicians. Set include_evidence to get the source behind each field. |
| count_practices | How many locations match a filter. Free, so an agent can size a query before it spends credits. |
| get_usage | The key's plan, credits used, and credits remaining this month. Free. |
A cited answer
The agent names the tool, then the source.
Evidence is what keeps a research agent from smoothing over a gap. If the field is unknown, the tool says so.
Start with one practice. Scale to all of them.
Get a key, read a record, and widen the query when the shape is right.