OpenAI WebMCP Challenge · Agent Discovery Copilot

Build an app while your agent discovers tools

Open influzer.ai inside ChatGPT’s browser (or Chrome with WebMCP enabled). This page exposes document.modelContext tools so you and your agent can search 358+ WebMCP websites and 14k+ classic MCP servers, inspect available tools, and open the best match — without leaving the build session.

Status: Starting… Mode: Detecting…

WebMCP Challenge Our listing Tool manifest

Try in ChatGPT (recommended for judges)

Open ChatGPT desktop → start a chat with the in-app browser.
Navigate to https://www.influzer.ai/webmcp/challenge
Ask: “Use recommend_agent_stack — I’m building a [your app]. Show WebMCP sites and MCP servers I should combine.”
Follow up: “Get tool schemas for the top WebMCP site and install steps for the top MCP server.”
Confirm, then: “Open the best WebMCP site in this tab.” (calls open_webmcp_site)

Example build-session prompts

“I’m building a booking app — recommend WebMCP sites for scheduling plus MCP servers for email/calendar.”
“Search MCP servers for Postgres and WebMCP sites with cart/checkout tools.”
“What tools does influzer.ai expose? Then navigate to the WebMCP directory.”
“Submit my site https://mysite.com for listing — my email is me@company.com.” (agent calls start_webmcp_listing_scan after you confirm)

Live tool console

Same tools ChatGPT discovers via getTools() — run them manually below.

Tool

Arguments (JSON)

Input schema

{}

Result

Run a tool to see structured output.

Log

Challenge entry — what’s new (Aug 25+)

recommend_agent_stack — dual-directory discovery for app building
get_mcp_server — full setup metadata for one classic MCP server
open_webmcp_site — navigate to a catalogued WebMCP site to use its tools
start_webmcp_listing_scan — submit a new WebMCP site (email + explicit user consent)
get_webmcp_listing_scan — poll scan progress and scorecard

Repo: github.com/rmencke2/logo · Docs: docs/WEBMCP_CHALLENGE.md