Use the planner from an agent

What it offers

OpenAPI

https://qrsntu.org/service/neurips26/openapi.json

llms.txt

https://qrsntu.org/service/neurips26/llms.txt

REST base URL

https://qrsntu.org/service/neurips26/api/v1

Full MCP

https://qrsntu.org/service/neurips26/mcp

Read-only MCP

https://qrsntu.org/service/neurips26/mcp/readonly

REST endpoints

Paths are relative to the base URL. The OpenAPI file has every parameter and response. A 429 response means slow down and retry.

MCP servers

Both use Streamable HTTP with plain JSON responses and need no authentication. Tool handlers are the same functions the REST API calls.

/mcp

All tools: list_conferences, search_events, get_event, get_schedule, find_overlaps, validate_itinerary, save_itinerary, get_itinerary, update_itinerary, export_calendar, get_venue_info, list_hotels.

/mcp/readonly

search and fetch only, for clients such as OpenAI deep research that expect those two tools.

Example: search, validate, save, open

Ask the user before saving. Event id 151174 stands in for an id from step 1.

BASE=https://qrsntu.org/service/neurips26/api/v1

# 1. Search. Each item has an id.
curl -s "$BASE/events?q=reinforcement+learning&city=sydney&date=2026-12-08&limit=5"

# 2. Validate a draft. Nothing is saved.
curl -s -X POST "$BASE/plans/validate" \
  -H 'Content-Type: application/json' \
  -d '{"title":"NeurIPS Sydney plan","city":"sydney",
       "items":[{"event_id":151174,"priority":"must"}]}'

# 3. Save. idempotency_key is a secret of at least 32 characters: use a new random UUID for each plan.
curl -s -X POST "$BASE/plans" \
  -H 'Content-Type: application/json' \
  -d '{"idempotency_key":"7c9e6679-7425-40de-944b-e07fc1f90ae7",
       "title":"NeurIPS Sydney plan","city":"sydney",
       "items":[{"event_id":151174,"priority":"must"}]}'

# 4. Open plan.url from the response. It looks like this:
#    https://qrsntu.org/service/neurips26/plan/{id}/
#    The response also has edit_token and edit_url. Keep them private. Repeating this request returns them
#    again only within 24 hours and before any update; reads never return them.

Connect your assistant

Try this prompt once connected:

Plan my NeurIPS Sydney attendance from December 6-12. Prioritize reinforcement learning, attention, agentic AI, and quantitative finance. Leave time for networking, avoid unnecessary scheduling conflicts, and generate a permanent itinerary link.

The steps follow each vendor's documentation as read on 1 Nov 2026. Model names are examples; use one your account has. See the status table for what has been tested.

Claude

claude.ai: add a custom connector with the URL https://qrsntu.org/service/neurips26/mcp.

Claude Code:

claude mcp add --transport http neurips26 https://qrsntu.org/service/neurips26/mcp

Messages API (MCP connector, beta):

curl https://api.anthropic.com/v1/messages \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "anthropic-beta: mcp-client-2025-11-20" \
  -d '{
    "model": "claude-opus-5-5",
    "max_tokens": 1000,
    "messages": [{"role": "user", "content": "Find NeurIPS Sydney talks on agentic AI."}],
    "mcp_servers": [{"type": "url", "url": "https://qrsntu.org/service/neurips26/mcp", "name": "neurips26"}],
    "tools": [{"type": "mcp_toolset", "mcp_server_name": "neurips26"}]
  }'

ChatGPT

Custom MCP server (formerly the developer mode connector): go to Plugins, select the plus button, then Add custom MCP server, and enter the URL above. OpenAI's help article says full MCP is for Business and Enterprise/Edu on ChatGPT web, and Pro users get read and fetch permissions only, so a Pro account can search but not save.

Responses API (MCP tool), request body. Approval is required unless you turn it off:

{
  "model": "gpt-6-astra",
  "tools": [{
    "type": "mcp",
    "server_label": "neurips26",
    "server_url": "https://qrsntu.org/service/neurips26/mcp",
    "require_approval": "always"
  }],
  "input": "Find NeurIPS Sydney talks on agentic AI."
}

GPT Actions: in the GPT editor, create a new action and import the schema from a URL: https://qrsntu.org/service/neurips26/openapi.json. OpenAI is retiring custom GPTs, with standard retirement scheduled for 11 December 2026, during the conference, and custom actions do not transfer automatically to plugins. Prefer MCP.

Gemini

Gemini CLI (MCP servers), in ~/.gemini/settings.json. Those docs say Antigravity CLI replaced Gemini CLI on 18 June 2026, so check the setup for your version:

{
  "mcpServers": {
    "neurips26": {
      "httpUrl": "https://qrsntu.org/service/neurips26/mcp"
    }
  }
}

Antigravity CLI: run agy mcp add neurips26 https://qrsntu.org/service/neurips26/mcp. In its settings file the key is serverUrl, not httpUrl.

Testing a local copy: start wrangler dev with --local-upstream localhost:<port>, otherwise links in responses point at qrsntu.org.

Gemini API, remote MCP (Interactions API), as a tools entry. Google's pages disagree on which models support it:

tools=[{
    "type": "mcp_server",
    "name": "neurips26",
    "url": "https://qrsntu.org/service/neurips26/mcp",
}]

Gemini API, function calling: declare the operations you need from openapi.json as functions, and call the REST API from your own code.

Gemini API, URL context (docs): add {"type": "url_context"} to tools and put https://qrsntu.org/service/neurips26/llms.txt in the prompt. This reads pages only, so it cannot save a plan.

Integration status

Tested on 1 Nov 2026 against a local copy of this server, using the example request above. Hosted clients need the public address and were not part of that run.

Data and privacy