Every score, benchmark, exercise, and tool here is callable by the assistant you already use. Nothing to install on our side. Nothing uploaded to us.
Direct answer: connect any AI assistant to this atlas over MCP at https://unitedwetransform.com/mcp, or point an agent that does not support MCP at /llms.txt.
The human side of pointing your AI at this atlas, deciding what to ask and what to do with the answer, is its own role. We call it the Cognitor.
Want one thing to paste, with no setup at all? Tell your assistant to read unitedwetransform.com/grade.md and give it your agenda. That single page carries the whole method: the eight pillars, the thirteen signals, and the template for a valid gathering.json. No MCP, no config, no account.
Agendas get drafted and rebuilt in a doc, a chat, an editor, a terminal, not a browser tab you have to remember to open. Building the atlas as a tool your assistant can call, instead of another site to visit, matches how the design work actually happens.
The Gathering Effectiveness Score is meant to be checked, not trusted blindly. That only works if the evidence is reachable from any client, in any assistant, with no login and no lock-in. We do not get to pick who is allowed to ask.
The grader and the team engine run the exact same in-process code whether you use the browser or an agent. What you send is never stored, never logged beyond an aggregate count, and never used to change the atlas.
Pick your client, copy the line, and the atlas is a tool your assistant can call.
claude mcp add --transport http uwt https://unitedwetransform.com/mcp
Run this in any project directory. The server appears the next time you start Claude Code.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
# ~/.codex/config.toml [mcp_servers.uwt] url = "https://unitedwetransform.com/mcp"
Add this block to config.toml, then restart Codex or start a new session. The ChatGPT desktop app's built-in Codex reads this same file, no CLI install needed.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
https://unitedwetransform.com/mcp
Settings, then Apps and Connectors, then Advanced, then Developer mode, then Create. Name it uwt, paste the URL, no auth. Deep research can also use this connector: the atlas exposes the search and fetch tools it requires.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
{
"mcpServers": {
"uwt": {
"url": "https://unitedwetransform.com/mcp"
}
}
}Paste into .cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
https://unitedwetransform.com/mcp
Settings, then Connectors, then Add custom connector. Paste the URL and click Add. No API key or OAuth needed.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
https://unitedwetransform.com/mcp
Cowork uses the same connector settings as Claude Desktop and claude.ai: Settings, then Connectors, then Add custom connector.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
{
"mcpServers": {
"uwt": {
"httpUrl": "https://unitedwetransform.com/mcp"
}
}
}Gemini CLI needs httpUrl for a remote streamable HTTP server. Its url key means something different (SSE) and will not connect here.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
{
"mcpServers": {
"uwt": {
"serverUrl": "https://unitedwetransform.com/mcp"
}
}
}Open Manage MCP Servers, then View raw config, and paste this block. Antigravity needs serverUrl, not the older httpUrl field.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
{
"servers": {
"uwt": {
"type": "http",
"url": "https://unitedwetransform.com/mcp"
}
}
}Paste into .vscode/mcp.json in your project. Open the Copilot Chat panel in Agent mode afterward to see uwt in its tool list.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
{
"mcpServers": {
"uwt": {
"url": "https://unitedwetransform.com/mcp"
}
}
}Paste this block into Windsurf's MCP config file, then refresh the MCP server list from Windsurf's settings panel.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
{
"mcpServers": {
"uwt": {
"url": "https://unitedwetransform.com/mcp"
}
}
}Open Cline's MCP Servers panel, choose Configure, then paste this same mcpServers JSON block.
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
https://unitedwetransform.com/mcp
Streamable HTTP, no auth, read-only. Discovery document (tool list, description): https://unitedwetransform.com/.well-known/mcp.json
If it fails: restart the client, then check its MCP/connector server list. Our endpoint always answers curl -X POST https://unitedwetransform.com/mcp with JSON, never a login page.
Any agent that can fetch a URL can still use this atlas. Point it at the navigation index first, then the plain JSON endpoints below.
curl https://unitedwetransform.com/api/statistics.json
curl https://unitedwetransform.com/api/score/{event-slug}.json
Ten tools, all read-only, served from the same MCP endpoint above.
| Tool | What it does | Example ask |
|---|---|---|
grade_agenda | Score a pasted agenda with the deterministic UWT rubric. Runs in-process; the text is never stored. | Grade this draft agenda. |
get_event_score | Fetch the cached scorecard for one of 23,624 scored events by slug. | What is the scorecard for this specific event? |
compare_events | Compare two cached scorecards pillar by pillar, with a one-line summary of which scores higher. | Compare two events I am considering sponsoring. |
search_gatherings | Search the gold-verified gathering index by free text, category, and minimum score. | Find the highest-scoring sales kickoffs in the gold set. |
search | Federated search across events, exercises, and core pages. Required by ChatGPT connectors and deep research. | Search the atlas for climate summit gatherings. |
fetch | Fetch the full content for an id returned by search. Required by ChatGPT connectors and deep research. | Get me the full scorecard for that event. |
get_statistics | Return corpus-wide summary statistics for the atlas. | What is the average score across all agendas? |
get_leaderboard | Return the gold-verified leaderboard, ranked by score. | Show me the top 10 highest-scoring events. |
get_exercises_for_pillar | Return the suggested improvement exercises for one scoring pillar. | Give me three exercises for weak follow-through. |
make_teams | Form balanced teams or breakout rounds from a pasted roster. Runs the same engine as /team-creator/; the roster is never stored. | Split these 60 attendees into balanced tables of 6. |
After connecting any client above, paste this prompt:
Use the uwt tools to show me the top 5 events on the leaderboard and grade this agenda: 9:00 Opening, 10:30 Workshop, 12:00 Lunch, 1:00 Panel, 4:00 Close.
A correct answer looks like a ranked list of 5 real event titles with scores in the high 40s to 50s, plus a grade for the pasted agenda somewhere under 40 with a note about missing mechanisms (this sample agenda has no named owners, no follow-up, and no stated problem). If the assistant instead says it cannot browse the web or has no tools available, the connection did not take: re-check the steps above.
Read-only. No accounts. No tracking of individual tool calls. The full privacy contract is at /agents.md. Found a score that looks wrong, or a tool that misbehaves? Email corrections@unitedwetransform.com, or use the correction link on any scorecard.