MCP Server
Use your organization's Pult data in AI assistants such as Claude, through the MCP server.
The Pult MCP server lets AI assistants read your organization's data in Pult. You ask a question in plain language, and the assistant fetches what it needs. It reads the same data as the REST API and can't change anything.
| What | Value |
|---|---|
| Server URL | https://api.pult.com/v1beta/mcp |
| Transport | Streamable HTTP |
| Authentication | OAuth, or an API key in the Authorization header |
What the assistant can see
- With OAuth, each person connects the assistant to their own Pult account, and it reads with that person's rights (see Who can connect).
- With an API key, the assistant reads with the key owner's admin rights, so it sees what an admin sees in the dashboard's reports. In a shared AI workspace (a Claude Team or Enterprise workspace, or a shared Langdock connection, for example) every member uses the same key, whatever their role in Pult. Create a dedicated key for each AI tool, so you can revoke it on its own.
See Authentication for both.
What you can ask
- "How many people were in each office location every day last week?"
- "On which days was Anna in the office this month?"
- "Who has an office booking at the Berlin office for Friday?"
- "Which team booked the most home office days in October?"
- "Which visitors are expected at the Berlin office tomorrow?"
Bookings, presence and visits cover at most 31 days per request. For a longer period, the assistant has to ask several times.
Tools
| Tool | What it returns |
|---|---|
list_users | Your organization's users, active ones by default |
get_user | One user, whatever their status |
list_groups | Your organization's groups |
get_group | One group |
list_offices | Your organization's offices, active ones by default |
get_office | One office, archived or not |
list_bookings | Bookings for up to 31 days |
get_booking | One booking |
list_presence_locations | The places where presence is detected |
get_presence_location | One presence location |
list_presence_days | Who was detected at which presence location, per day, for up to 31 days |
list_presence_location_counts | How many people were detected per presence location and day, for up to 31 days |
list_visits | Visits for up to 31 days, in every status |
get_visit | One visit |
Each tool matches one endpoint of the reference and returns the same data, except that characters that hide or reorder text (such as zero-width spaces and direction controls) are removed, so the assistant reads what you would see. The REST API returns text exactly as stored.
The assistant sees only the tools your key or connection, and your organization, can use: the booking tools need Pult Workplace, the office tools Pult Workplace or Pult Visitors, the presence tools Pult Presence, and the visit tools Pult Visitors. The server also names the tools it left out and why, so an assistant that receives this can tell you what to do.
Errors
Rate limits and errors are the same as for the REST API. They reach you in one of two ways:
- As the tool's result, when a call is refused with
organization_disabled,api_not_enabled,api_key_paused,connection_paused,feature_not_enabledorrate_limited. The result carries the same error object as the REST API, so the assistant can explain it, or wait the seconds indata.retryAfterSecondsand try again. - As an HTTP error to your AI tool, for a missing, invalid, expired or revoked key or access token, a key or connection that lacks a tool's scope, the per IP address limit, and any error while your AI tool connects. These reach your AI tool rather than the assistant. See Authentication for these errors and Rate limits for the wait.
Connect an AI tool
With OAuth
Add the server URL in your AI tool, without a key. If the tool asks how to authenticate, choose OAuth, and if it offers to send the resource parameter, leave that on. The tool opens Pult in your browser: sign in and approve the connection (see Connect with OAuth).
| AI tool | Status |
|---|---|
| Claude | Tested |
| ChatGPT | Tested |
| Claude Code | Tested |
| Langdock | Tested |
| Microsoft 365 Copilot | Tested |
| Cursor | Expected to work |
| VS Code | Expected to work |
"Tested" means we connected the tool to Pult end to end. "Expected to work" means the tool's own documentation says it registers itself and sends the resource parameter, and we have not tested it yet. Other tools that do both should work too, whether they register with a client secret or without one.
Microsoft 365 Copilot: in the Microsoft 365 admin center, connect Pult's MCP server with DCR (OAuth 2.0). Then select Authorize right away, sign in to Pult and approve as an admin. Pult may remove a registration that never connects an account, so don't leave it for later. In a tenant without a Copilot license, the admin center can stop with "an access token could not be acquired for the authentication service" before it reaches Pult: connect from a tenant with Copilot.
With an API key
Add the server URL in your AI tool and send your key in the Authorization header. Tools ask for
the key in one of two ways:
- The whole header value: type
Bearer pult_<key>, with the space. - Only the key: type
pult_<key>. The tool addsBeareritself, so typing it would double it.
| AI tool | What to enter | Status |
|---|---|---|
| Claude Code | --header "Authorization: Bearer pult_<key>" | Tested |
| Claude | Request header authorization: Bearer pult_<key> | Expected to work |
| Copilot Studio | API key, type Header, Authorization: Bearer pult_<key> | Expected to work |
| Langdock | API key, header type "Authorization: Bearer": pult_<key> | Expected to work |
| Le Chat | API token, Bearer: pult_<key> | Expected to work |
| n8n | Bearer Auth credential: pult_<key> | Expected to work |
| Codex | Bearer token: pult_<key> | Expected to work |
| Cursor | Header Authorization: Bearer pult_<key> | Expected to work |
| VS Code | Header Authorization: Bearer pult_<key> | Expected to work |
| Gemini CLI | Header Authorization: Bearer pult_<key> | Expected to work |
| Zed | Header Authorization: Bearer pult_<key> | Expected to work |
"Expected to work" means the tool's own documentation supports an API key in a header, and we have not tested it yet. In Claude, the Request headers option is a beta that not every account has yet.
For Claude Code, add the server with one command:
claude mcp add --transport http pult https://api.pult.com/v1beta/mcp \
--header "Authorization: Bearer pult_<key>"Not supported yet
Gemini Enterprise connects to MCP servers only with a client ID and secret that an admin registers with the server in advance. Pult doesn't offer registering apps that way yet.
Last updated on
List visits GET
Lists the visits whose stay overlaps a window of at most 31 days, read in each visit's own local dates, in every status. The order is stable while you follow `nextCursor` but otherwise unspecified and may change; sort by `date` or `startsAt` on your side. Each visitor is one visit: a group visit lists one visit per visitor. A visit with no end, or with an end at or before its start, counts as a one-day stay on its start date. Visits your retention setting deleted no longer appear. Moving a calendar meeting to another calendar cancels its visit and creates a new one with a new id. Visits from before 2026-10-27: no-shows were inferred later, and cancelled ones are missing. Re-read a recent window to see changes.
Changelog
Every change to the Pult API, newest first.