MCP server

Pav runs a remote Model Context Protocol server. Every read endpoint of the API is an MCP tool, so an agent can search programs, trials, deals, patents and FDA records and cite the records it used.


URL	`https://api.pav.bio/mcp`
Transport	Streamable HTTP
Auth	Your Pav API key: `Authorization: Bearer <api_key>`. See Authentication.
Access	Read-only. Every tool is marked read-only.
Tools	18, one per API endpoint

Connect a client

Claude Code
Claude Desktop
Cursor
Python

claude mcp add --transport http pav https://api.pav.bio/mcp \
  --header "Authorization: Bearer $PAV_API_KEY"

Verify with claude mcp list, or type /mcp inside Claude Code.

Claude Desktop reaches remote servers that need a header through the mcp-remote adapter (requires Node.js 18+). Open Settings → Developer → Edit Config and add:

claude_desktop_config.json

{
  "mcpServers": {
    "pav": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.pav.bio/mcp",
        "--header",
        "Authorization:${PAV_AUTH_HEADER}"
      ],
      "env": {
        "PAV_AUTH_HEADER": "Bearer <your_api_key>"
      }
    }
  }
}

Restart Claude Desktop. The Pav tools appear under the tools menu in a new chat. Keep Authorization:${PAV_AUTH_HEADER} without a space after the colon; the value comes from env.

Add to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project):

mcp.json

{
  "mcpServers": {
    "pav": {
      "url": "https://api.pav.bio/mcp",
      "headers": { "Authorization": "Bearer <your_api_key>" }
    }
  }
}

Check Settings → MCP shows pav with its tools.

Use the FastMCP client: pip install fastmcp.

import asyncio
import os

from fastmcp import Client
from fastmcp.client.transports import StreamableHttpTransport

transport = StreamableHttpTransport(
    "https://api.pav.bio/mcp",
    headers={"Authorization": f"Bearer {os.environ['PAV_API_KEY']}"},
)


async def main():
    async with Client(transport) as client:
        tools = await client.list_tools()
        print(len(tools), "tools")
        result = await client.call_tool(
            "list_deals",
            {"min_value_usd": 1_000_000_000, "sort": "headline_value", "limit": 3},
        )
        for deal in result.structured_content["data"]:
            print(deal["deal_name"], "|", deal["total_value"]["raw_text"])


asyncio.run(main())

Output

18 tools
Pfizer acquisition of Seagen | approximately $43 billion
Amgen acquisition of Horizon Therapeutics plc | transaction equity value of approximately $27.8 billion
Abbott acquisition of Exact Sciences | enterprise value of ~$23 billion

Any other MCP client that supports streamable HTTP and custom headers works with the same URL and header.

Tools

Each tool is one API endpoint. Its arguments are the endpoint’s parameters and its result is the endpoint’s JSON response. The dataset pages describe the fields and filters.

Tool	Endpoint	Use
`list_programs`	`GET /v1/programs`	Search and filter programs.
`get_program`	`GET /v1/programs/{program_id}`	One program with trials and ontology terms.
`get_similar_programs`	`GET /v1/programs/{program_id}/similar-programs`	Competitors and analogues of a program.
`list_companies`	`GET /v1/companies`	Resolve a company name to `company_id`.
`get_company`	`GET /v1/companies/{company_id}`	One company.
`list_company_programs`	`GET /v1/companies/{company_id}/programs`	A company’s pipeline.
`list_company_trials`	`GET /v1/companies/{company_id}/trials`	Trials linked to a company.
`list_company_changes`	`GET /v1/companies/{company_id}/changes`	A company’s pipeline changes.
`list_trials`	`GET /v1/trials`	Search and browse clinical trials.
`get_trial`	`GET /v1/trials/{nct_id}`	One trial.
`list_deals`	`GET /v1/deals`	Search and filter deals.
`get_deal`	`GET /v1/deals/{deal_id}`	One deal with terms, events and sources.
`list_patents`	`GET /v1/patents`	Search and filter patent families.
`get_patent_family`	`GET /v1/patents/{family_id}`	One family with members, ownership and term.
`list_fda_records`	`GET /v1/fda`	Search Orange Book, Purple Book, orphan, warning-letter and recall records.
`get_fda_record`	`GET /v1/fda/{record_key}`	One FDA record.
`list_changes`	`GET /v1/changes`	The pipeline change feed.
`get_stats`	`GET /v1/stats`	Dataset counts by phase and modality.

Resolve a company name with list_companies first, then pass its company_id to list_company_programs, list_deals, list_patents, list_fda_records or list_changes.

Example prompts

“Which KRAS G12C inhibitors are in Phase 3, and who owns them?”
“List recruiting Phase 3 obesity trials and their sponsors.”
“What are the five largest biopharma acquisitions by headline value? Cite the source filing for each.”
“Which live US patent families does Eli Lilly own that mention tirzepatide, and when do their statutory terms end?”
“List the Orange Book patents and exclusivities for Mounjaro and when each expires.”
“What changed in Eli Lilly’s pipeline since June 1?”
“Find programs similar to Merck’s calderasib.”

Errors and limits

A missing or invalid key returns HTTP 401 before any MCP message is handled.
An API error inside a tool, such as an unknown deal_id, returns an MCP tool error that carries the API’s { error: { code, message, request_id } } body.
Each tool call counts as one request toward your key’s rate limit. Connecting and listing tools do not count.

Get started

Datasets

Working with the API

Integrations

Guides

Connect a client

Tools

Example prompts

Errors and limits

​Connect a client

​Tools

​Example prompts

​Errors and limits

Connect a client

Tools

Example prompts

Errors and limits