Living reference

Connect your AI to Pulse (MCP)

Pulse is an MCP server: your own AI (Claude Code, Claude Desktop, or the API's MCP connector) connects to a workspace and builds & reads events directly. No AI runs inside Pulse — you bring the agent.

1. Get a token

In the Pulse console, open Connect AI and create a personal access token. It's shown once (pulse_pat_…). Give it Read only or Build & edit events, and an expiry.

The Connect AI panel in the Pulse console: a token mint form with a name, a capability (Build & edit events / Read only) and an expiry, above ready-to-paste config for Claude Code, Claude Desktop, and the API MCP connector. — Console → **Connect AI**: mint a token, then copy the config for your client.

The reveal-once card showing a freshly minted pulse_pat_ token with a Copy button — Pulse stores only its hash, so it can't be shown again. — The raw token is shown **once** — copy it now; Pulse keeps only its hash.

2. Point your client at `/mcp`

The endpoint is https://<your-workspace-host>/mcp, authenticated with Authorization: Bearer pulse_pat_…. The console gives you ready-to-paste config for the CLI, Claude Desktop, and the API connector.

claude mcp add --transport http pulse https://nowpresenting.net/mcp \
  --header "Authorization: Bearer pulse_pat_…"

The boundary

Setup-time write + read-anytime. An agent can create and edit draft events and read reports. It cannot run the live room (advance slides, launch polls, moderate) and cannot delete — the room is human-run. A read-only token can't write at all.

Tokens & limits

Every token you mint lists in the console with its capability, expiry, and last-used time; revoke any of them there. A token's capability can never exceed the role of the person who minted it.

The tokens table in Connect AI listing minted tokens with their capability, created and expiry dates, last-used time, and a revoke action. — Mint, review, and revoke tokens from the console.

Calls are rate-limited per token (a burst returns 429 with Retry-After), and an operator can disable MCP workspace-wide (calls then return 503). Both are transient — retry after the window.

Tools

Tool	Access	What it does
`whoami`	read	Confirm the connection (workspace, role cap).
`list_sessions`	read	Events in the workspace (id, slug, title, status).
`get_session`	read	One event's status, links, and content counts.
`export_blueprint`	read	An event as an editable blueprint JSON.
`validate_blueprint`	read	Check a blueprint; returns `{ok}` or `{path,error,hint}` errors.
`get_report`	read	The end-of-event report (ended events only).
`create_session_from_blueprint`	write	Create a new draft event; returns id + join/setup links.
`update_session_from_blueprint`	write	Re-apply a blueprint onto a draft in place (id/join unchanged).

Resources & the prompt

pulse://spec/event-blueprint — the live blueprint schema (read it before authoring).
pulse://spec/mcp — this reference, as an MCP resource.
The build-my-event prompt walks the review → interview → validate → create flow.

The authoring loop: read the schema resource, draft a blueprint, validate_blueprint, fix each error by its path, then create_session_from_blueprint. It's the same one blueprint format used everywhere in Pulse.

Not yet

claude.ai web custom connectors (which require OAuth) aren't supported yet — the token flow works with Claude Code, Claude Desktop, and the API MCP connector today.

Back to the FAQ · the Claude skill