Surveys with AI agents (MCP)

The Gleap MCP server lets AI agents such as Claude, ChatGPT, Cursor or Kai work with your surveys: draft a survey from a one-line request, check its logic, read results and AI insights, and draft follow-ups to respondents. Agents create and edit drafts; publishing and sending stay with your team.

Connect

Add the Gleap MCP server as a custom connector in your AI client and sign in with Gleap (OAuth):

Data region	MCP server URL
EU (default)	`https://mcp.gleap.io/mcp`
US	`https://mcp.us.gleap.ai/mcp`

Clients that can’t do OAuth send the headers x-gleap-api-key and x-gleap-project-id instead. The agent acts with your role in the project.

Tools

Tool	What it does	Access
`list_surveys`	Surveys with id, SDK id, status, format, published version and unpublished changes. `get_surveys` is an alias.	Read
`get_survey`	The draft (or published) definition — questions, logic, endings, hidden fields — plus page settings and targeting.	Read
`create_survey`	New survey from a definition, optionally with an in-app event trigger and audience. Always a draft.	Draft only
`update_survey`	Replace the draft definition. On a live survey it saves the next draft; respondents keep seeing the published version.	Draft only
`validate_survey`	Check a definition or the stored draft: loops, unreachable questions, dead ends, conditions that don’t fit the question type. Saves nothing.	Read
`publish_survey`	Publish the draft through Gleap’s publish endpoint. Agents get `requiresApproval` — a teammate publishes.	Needs approval
`pause_survey`	Take a live survey off.	Write
`get_survey_results`	Totals, start and completion rates, NPS, per-question counts and drop-off, endings, channels and common paths — from stored counters — plus the question funnel when analytics events exist.	Read
`get_survey_analytics`	Analytics with the previous period for comparison: views, starts, submissions, rates, the question funnel with drop-off, timing, score, versions, device and channel breakdowns, busiest times, reminders, email and quality counts. Filters: date range, version, channel, device.	Read
`get_survey_insights`	AI themes with sentiment, summary, drivers, segments and spike alerts. Returns the stored insights at once; after new responses, reading them starts one update in the background (`updating: true`, read again in a minute).	Read
`ask_survey_results`	A plain-language question → an answer with cited response ids. Uses AI (token usage).	Read
`get_survey_responses`	Compact response rows. Filters: status, answer by key (`{ "nps_score": "0..6" }`), theme, channel, contact, text. Cursor pagination.	Read
`get_survey_response`	One response with labelled answers, the path taken, ending and contact.	Read
`reach_out_to_respondent`	Start a messenger or email conversation linked to a response. Returns a draft by default; sends only with `send: true`.	Needs approval
`get_survey_summary`	Stored AI summary of a survey’s responses (legacy).	Read

How agents build a survey

create_survey with a definition. The tool rejects common mistakes before saving — unknown properties, duplicate ids or keys, jumps to unknown or earlier questions, conditions on choice labels instead of choice ids, NPS ranges outside 0–10 — and returns Gleap’s validation of the saved draft.
The agent fixes problems with update_survey (always the complete definition) or validate_survey.
You open the reviewUrl the tool returns, check the survey in the editor and click Publish changes.

Example request to an agent:

Create a churn survey for customers who cancel. Ask why, and if it’s pricing, ask what price would have worked. Show it full screen right after they cancel.

The agent creates a full-screen draft with a single question, a jump from the too_expensive choice to a number question, two endings and the trigger subscription_cancelled (once per contact) — and tells you it hasn’t published it.

Guardrails

create_survey always saves status: "draft"; update_survey only writes the draft definition. Neither publishes nor changes a live survey’s version, status or targeting.
publish_survey goes through the same publish endpoint as the dashboard. For agents it returns { "published": false, "requiresApproval": true, "reviewUrl": "…" }; the agent shouldn’t retry.
reach_out_to_respondent sends nothing unless the call passes send: true, which agents should only do after you approved the exact message.
Results stay compact: lists are capped, long answers are clipped and responses are paged with a cursor.
Customer-facing agent runs (Kai answering one customer) can’t use survey tools.

Get Started

SDKs

Kai Code

Surveys

Server-side event tracking

Guides

Connect

Tools

How agents build a survey

Guardrails

​Connect

​Tools

​How agents build a survey

​Guardrails

Connect

Tools

How agents build a survey

Guardrails