MCP Server

Connect AI assistants to Sessionboard via Model Context Protocol

Early Access — The MCP server is available to organizations with AI features enabled. Reach out to your Sessionboard account manager if you’d like it turned on.

The Sessionboard MCP (Model Context Protocol) server lets AI assistants like Claude, ChatGPT, Microsoft Copilot, and Google Gemini query and manage your event data directly. Built on the open Model Context Protocol standard, it exposes 39 tools covering search, reads, natural-language queries, cited answers about an event, reports and dashboards, and write operations.

This page is the technical reference. If you just want to connect an assistant, the Help Center has friendlier step-by-step guides with screenshots — start at the AI assistants overview:

• Connect Claude — including Sessionboard’s listing in Claude’s connector directory
• Connect ChatGPT — via a developer-mode connector
• Connect Microsoft Copilot — via a Copilot Studio agent
• Connect Google Gemini — Gemini app, Gemini Enterprise, or the Gemini CLI

What is MCP?

Model Context Protocol is an open standard that enables AI assistants to securely connect to external data sources and tools. Instead of copying data into prompts, MCP lets the AI query and update your data on demand through a structured interface.

Server URLs by region

Each Sessionboard region runs its own MCP server. Use the one matching where your organization is hosted:

Region	MCP server URL
United States	https://mcp.sessionboard.com/mcp
Europe	https://mcp-eu.sessionboard.com/mcp
Middle East	https://mcp-me.sessionboard.com/mcp

The server uses the streamable HTTP transport and supports OAuth 2.0 with PKCE and dynamic client registration — most clients discover the sign-in flow automatically.

Prerequisites

Two things must be true before the hosted MCP server will accept a sign-in:

1. AI features are enabled for the organization. Contact Sessionboard support or your account manager if they are not.
2. The person signing in has the Connect AI assistants permission. Organization administrators have it. Other roles need it granted in role settings.

Claude, ChatGPT, Copilot, Gemini, and any other remote client connect with OAuth. There is no MCP switch on the organization or on an API token, and pasting an API token into the assistant will not authenticate. The hosted server (https://mcp.sessionboard.com/mcp and the regional hosts) accepts the OAuth bearer token from that sign-in.

An API token is for scripts calling the public API with x-access-token. Those calls to /v1/insights/... need the read:insights scope (a token with an empty scope list is the legacy full-access case and can call them too). Write scopes such as write:sessions are separate.

Setup

1. Connect your AI client

   Most assistants (Claude, ChatGPT, Copilot Studio, Gemini) connect through their own UI — follow the Help Center guides above. For clients configured by file (Claude Code, Cursor, and similar), add:

   {
     "mcpServers": {
       "sessionboard": {
         "url": "https://mcp.sessionboard.com/mcp"
       }
     }
   }
   

   Swap in the EU or Middle East URL from the table above if that’s where your organization is hosted.

2. Sign in

   On first use the client walks you through Sessionboard’s OAuth sign-in and organization consent automatically — no token to copy. A client that cannot complete that sign-in cannot use the hosted MCP URL. Scripts call the public API with an API token instead.

3. Start querying

   Ask your AI assistant questions about your event data in natural language. The assistant uses the MCP tools to query Sessionboard and return results.

Available Tools (46)

The MCP server exposes 46 tools in five groups. Write operations use a manage_* pattern — one tool per entity with an action parameter to pick the operation (e.g., action: "create").

All tools include MCP annotations for safe AI behavior — read tools are marked readOnlyHint: true, every manage_* tool is marked destructiveHint: true because each can overwrite an existing record through its update action (even the ones with no delete), and all are openWorldHint: false because every tool is bounded to your organization’s data. Clients such as Claude and ChatGPT use these hints to ask for confirmation before a write runs.

Search & Discovery (11 tools)

Find events, search your data, ask questions in plain English, and explore your schema.

Tool	Description
list_events	List all events in the organization
find_event	Search events by name
get_event_overview	Event summary — session counts, speakers, tracks, status breakdown
search	Full-text search across sessions and speakers with citation URLs
fetch	Fetch a single search result by ID (pairs with search)
get_schema	Browse your data model — entities, fields, and relationships
query_data	Ask a question in plain English and get results (AI writes the query for you)
execute_sbql	Run a SbQL query directly with pagination
get_suggestions	Get suggested queries based on your data
get_widget_schema	Widget types, configs, and grid layout rules for dashboards
ask_event	Ask a question about one event and get a cited answer — see Asking about an event. Needs read:ask

Read Details (4 tools)

Fetch sessions and contacts. All read-only and paginated.

Tool	Description
list_sessions	List sessions for an event (paginated, up to 100 per page)
get_session	Get a single session by ID with full details
list_contacts	List contacts for an event (paginated, up to 100 per page)
get_contact	Get a single contact by ID with full details

Read-Only Lists (9 tools)

Read counterparts to the manage_* tools — safe to call without any write scopes.

Tool	Description
list_metadata	Rooms, tracks, tags, formats, levels, languages, statuses, and custom fields
read_agenda_drafts	List and inspect agenda drafts
list_draft_sessions	Session placements within a draft
read_event_rules	Scheduling rules and constraints
read_personas	Evaluation personas
read_event_dashboards	Event-scoped dashboards
list_event_reports	Event-scoped saved reports
read_org_dashboards	Org-level dashboards (cross-event)
list_org_reports	Org-level saved reports (cross-event)

Create & Manage (15 tools)

Each manage_* tool handles all operations for an entity — create, update, delete, and more — using an action parameter. For example, manage_session with action: "create" creates a session, while action: "update" updates one.

Tool	Actions	Scope	Description
manage_metadata	list, create, update	write:metadata / write:fields	Rooms, tracks, tags, formats, levels, languages, statuses, and custom fields
manage_session	create, update, restore, update_fields	write:sessions	Session CRUD + custom field values
manage_contact	create, update, restore	write:contacts	Contact CRUD (find-or-create by email)
manage_exhibitor	create, update, restore	write:exhibitors	Exhibitor CRUD
manage_sponsor	create, update, restore	write:sponsors	Sponsor CRUD
manage_agenda_draft	list, get, create, update, delete, get_changes, commit	write:events	Agenda draft lifecycle
manage_draft_sessions	list, create, update, delete, bulk	write:events	Session placements in drafts
manage_event_rules	list, get, create, update, delete	write:events	Scheduling rules and constraints
manage_personas	list, get, create, update, delete	write:events	Evaluation personas
manage_event_dashboard	list, get, create, update, delete	write:events	Event-scoped dashboards
manage_event_widget	create, update, delete	write:events	Event dashboard widgets
manage_event_report	list, run, create, create_from_nl, update, delete	write:events	Event-scoped saved reports
manage_org_dashboard	list, get, create, update, delete	write:events	Org-level dashboards (cross-event)
manage_org_widget	create, update, delete	write:events	Org-level dashboard widgets
manage_org_report	list, run, create, create_from_nl, update, delete	write:events	Org-level saved reports (cross-event)

Registration (7 tools)

Read-only answers about an event’s registration: the funnel, attendees, orders, payments, membership, capacity and setup. They need the read:registrations scope, which is in preview: ask Sessionboard support to add it to your token, or name it in your OAuth scope request. The event must have Registration turned on (Early Access); otherwise these tools say so instead of answering.

Tool	Actions	Description
read_registration_summary	—	Registrations by status, by day, per ticket type, attendee type and form; member mix; revenue; orders per status; invoices still owed; capacity pools
find_registrations	search, get	Find attendees by a name, email or company fragment, or filter by status, ticket type, attendee type or form; one registration in full
read_registration_orders	list, get	Find an order by number or buyer email, or list orders by status (awaiting_invoice, partially_paid still owe); one order in full
read_registration_payments	list, get	Payment records: card and bank payments through your payment gateway and offline money, failed attempts with their reason, refunds and the charge they refund. Never card numbers or tokens
read_registration_members	lookup, list	Check one email the way the form does (member, lapsed or none), or list registrations with the membership recorded when they registered
read_registration_capacity	—	Every ticket type’s capacity, sold and available, plus shared capacity pools
read_registration_setup	section	Fee groups and their promo codes, donation funds with raised and pledged totals, workshop picking rules, the badge print queue

Try: “How many people registered for Annual Meeting 2026 last week, and how much is still owed on invoices?” or “Is [e-mail] a member, and is she registered?”

Asking about an event

ask_event answers a question about one event the way an organizer would want it answered. The answer is grounded in the event’s program (sessions, speakers, exhibitors, sponsors), session transcripts, the event’s knowledge and pinned answers, and the Sessionboard Help Center, and it cites its sources with [n] markers listed under Sources. It calls POST /v1/event/{eventId}/ask.

Input	Required	Description
event_id	Yes	The event to ask about
question	Yes	The question, in plain language (up to 500 characters)
conversation_id	No	The conversation_id from a previous answer, to ask a follow-up in the same thread
lang	No	Answer language as a language tag (e.g. en, fr, pt-BR)

• Scope — needs read:ask. Assistants connected before this scope existed hold only the scopes they were granted; reconnect to grant it.
• Read-only — marked readOnlyHint: true. Ask never changes the event.
• Use it for synthesis — “which sessions cover X”, “what did speakers say about Y”, “how do I set up Z”. For exact counts and lists, execute_sbql and query_data are faster and exact.
• Slow — an answer takes several seconds; the tool waits up to 90 seconds.
• Follow-ups — the answer ends with a conversation_id. Pass it back to continue the thread.
• Refusals come back as tool errors — Ask turned off for the event, AI features off, the per-user question limit (with the seconds to wait), the event’s daily Ask budget used up, or a question Ask would not take.

Parameter Conventions

• Event scoping — All event-scoped tools require event_id (snake_case string).
• Action selection — Consolidated tools use action to pick the operation (e.g., action: "create").
• Entity IDs — snake_case: session_id, contact_id, entity_id, draft_id, widget_id, query_id.
• Optimistic concurrency — Update actions require updated_at (ISO 8601 timestamp from the last read). Returns 409 Conflict if the entity was modified since.

Result Size Safety

All results are automatically truncated to prevent context window overflow:

• List results — Capped at 50 items. When truncated, the response includes truncated: true, total_count, and a hint with guidance on pagination.
• Query results — Capped at 100 rows with truncation metadata.
• SbQL queries — execute_sbql and query_data default to a 50-row limit. Use the limit parameter for more.

Resources

The MCP server also exposes two MCP resources that AI clients can load into context:

URI	Description
sbql://schema	The SbQL schema — entities, fields, relationships, and format functions
sbql://events	List of events in the organization with metadata

Prompt Templates

Pre-built templates that structure the AI’s approach to common report types:

Prompt	Description
speaker_roster	Generate a speaker roster with contact info and session assignments
session_schedule	Generate a day-by-day session schedule grid
evaluation_summary	Generate an evaluation/review score analysis
task_status	Show outstanding tasks across speakers and exhibitors

Security

• Org isolation — The SbQL compiler injects org_id into every query from the authenticated token. Prompt injection cannot access other organizations’ data.
• PII masking — OAuth connections always mask PII in query results: emails become j***@a***.com, phone numbers become ***-***-4567. For API-token calls to the query routes, masking follows that token’s PII Obfuscation setting (off by default — turn it on for any token a third party will use). See privacy details in the Help Center.
• Scoped access — Tokens are restricted to specific scopes and optionally to specific events.
• Audit logging — Every API call is recorded with source, method, response time, and token ID. For ask_event the question text is redacted from the log.

Example Conversations

Once connected, you can ask questions like:

• “How many sessions are accepted for this year’s conference?”
• “Show me the top 10 speakers by number of sessions.”
• “What’s the breakdown of sessions by track and status?”
• “List all sessions that don’t have a room assigned yet.”
• “Which sessions cover cardiac imaging, and what did the speakers say about it?”
• “Create a new track called ‘AI & Machine Learning’ with color #4F46E5.”
• “Build a dashboard showing session counts by status and track.”
• “Create a draft agenda and schedule the keynote in the Main Hall at 9am.”