Test whether an AI agent can actually complete a website conversion journey.
Give an MCP-capable agent a public website and a commercial task. Agent Conversion runs a real browser journey and returns the observed outcome and evidence—not just a readiness score.
Use it for pricing discovery, starting a self-service signup, or reaching a demo/contact-sales path.
Connect to the remote endpoint
https://agent-conversion-coral.vercel.app/api/mcpStreamable HTTP; no API key required. Audit starts are rate-limited. Connecting and listing tools do not start an audit.
Two tools, one workflow
Connect → test_conversion → receive audit_id → get_conversion_report → inspect outcome and evidence
test_conversion — START a browser audit
Pass url (a public HTTP(S) website), task (the goal in plain language), and optionally locale (en or es). The audit starts asynchronously. The response includes a public audit_id, pending status, a report URL and retry_after_seconds. Call this once per task; do not restart it while waiting.
get_conversion_report — POLL / RETRIEVE the result
Pass the returned audit_id. If pending, wait retry_after_seconds and poll again. A completed result includes task_completed, evidence steps, a summary, recommendations and a score when the run supports one. An incomplete or infrastructure-failed run can have its score withheld.
Minimal example task
Find the monthly price of the Pro plan on https://www.example.com/pricing.Copyable prompt for an MCP-capable agent
Use Agent Conversion to test whether an AI browsing agent can find the monthly price of the Pro plan on https://www.example.com/pricing. Call test_conversion once, then poll get_conversion_report with its audit_id until it finishes. Tell me whether the task was completed, cite the evidence steps, distinguish a site refusal from an incomplete or infrastructure run, and include the report URL. Do not sign up, submit personal data, or buy anything.The example URL is illustrative. Replace it with a public site you are authorized to test before starting an audit.
Connection examples
These snippets only connect and discover tools; they do not run an audit. Use your agent with the prompt above when ready.
FastMCP (Python)
import asyncio
from fastmcp import Client
async def main():
async with Client("https://agent-conversion-coral.vercel.app/api/mcp") as client:
print([tool.name for tool in await client.list_tools()])
asyncio.run(main())Official TypeScript MCP client v2
import { Client, StreamableHTTPClientTransport } from "@modelcontextprotocol/client";
const client = new Client(
{ name: "my-client", version: "1.0.0" },
{ versionNegotiation: { mode: "auto" } },
);
await client.connect(new StreamableHTTPClientTransport(new URL("https://agent-conversion-coral.vercel.app/api/mcp")));
console.log((await client.listTools()).tools.map((tool) => tool.name));
await client.close();Pydantic AI
from pydantic_ai.mcp import MCPToolset
toolset = MCPToolset("https://agent-conversion-coral.vercel.app/api/mcp")
# Attach to your configured Agent(..., toolsets=[toolset]).Attach this toolset to an Agent configured with your own model provider and credentials; the MCP endpoint itself needs no key.
What to expect
This is an early public capability. Audits execute real browser journeys and may take a few minutes. Reports separate the observed task outcome from site-facing access refusal and from runs that did not finish or had infrastructure trouble, where supported by the evidence. The agent never buys, pays or submits real personal data.