121surveys
Sign in
Menu

Claude integration

Create and analyse surveys with Claude – through the 121surveys MCP server

121surveys provides an MCP server (Model Context Protocol) at https://www.121surveys.com/mcp. Through it, Claude and other AI assistants reach most of the editor and the analysis: building questionnaires including filter logic and research methods, testing every path, reading results and simulating conjoint markets – with exactly the permissions and role of the person who granted access.

Building and testing is free. To publish, you choose a paid plan.

What Claude can do in 121surveys via MCP

  • Build the questionnaire

    Claude creates a draft, adds pages and inserts questions one by one or as a batch – all or nothing, with one error message per invalid entry. Questions can be edited, moved and duplicated; on top come link variables, actions after submission and the question pool with randomly drawn questions.

  • Filter logic and research methods

    Claude sets page conditions, branching with jumps or survey end, and conditions on individual questions just as in the editor; invalid logic is rejected with a reason. For market research there are dedicated method questions, among them conjoint, menu-based conjoint, MaxDiff, brand-price trade-off, Van Westendorp, Gabor-Granger, Kano, claims test, image A/B test and heatmap.

  • Test run of every path

    Before publishing, Claude runs the same check as the publish dialog and creates test submissions: one for each possible route through the survey, or a chosen number with random answers. The response lists, for each path, the pages, the steering answers and how it ends. Test data counts in no statistics.

  • Results and statistics

    Claude reads key figures, answers per question, crosstabs with a chi-square test, NPS analysis, trends, correlations, the quality report and quotas – with the same numbers as the app’s analysis. Each method question has its own results tool, and raw data comes as a short-lived, signed export link.

  • Conjoint simulation and portfolio

    The market simulator calculates shares of preference using the logit rule for products you define, including against the “none of these” option. From portfolio scenarios with rules, unit costs and a minimum margin, Claude finds the best product line-up – by contribution margin or reach, with an uncertainty interval over the hierarchical Bayes draws.

  • Distribute and read credits

    Claude creates distribution links and one-time links, and sends personal invitations and reminders to everyone who has not finished yet. get_credits shows available credits, pots with expiry dates and waiting responses; credits cannot be bought via MCP.

Connect in three steps

  1. Enter the address

    In claude.ai, go to the settings and add a custom connector with the address https://www.121surveys.com/mcp; ChatGPT takes the same address as a connector. Claude Desktop, Claude Code and your own programs connect with an API key that people with the Administrator role create under Manage Workspace → API, MCP & webhooks.

  2. Grant access

    When connecting via OAuth, you sign in in the browser, choose the workspace and confirm the permissions – no key needed. The access token is short-lived and is renewed automatically. With a key, you enter it as a bearer token in the configuration, for Claude Desktop via the mcp-remote package.

  3. Check the connection

    Ask Claude to call whoami: the reply names the person, workspace and granted permissions. When new tools are added, disconnect the connector and connect it again; with a key, restarting the app is enough. n8n and Make use the same key to talk to the REST API.

Permissions, roles and the delete lock

Which permissions does the assistant get?

Keys and OAuth share the same four scopes: read, write, admin and delete. Only people with the Owner or Administrator role of the account can grant admin and delete; for any other member, the consent page grants the rest and explains why. You disconnect your own connections under My account → Connected apps, and access ends immediately.

Can Claude do more than I can in the app?

No. The assistant acts as the person who granted access. Every tool requires the same minimum role as the part of the app that does the same thing – from Viewer to Administrator. tools/list only shows what the role allows, and the role is re-read before every call so that a role change takes effect immediately.

Can Claude delete surveys?

Only behind a triple lock: the access carries the delete scope, it belongs to a person with the Owner or Administrator role of the account, and the switch “Allow deleting via AI assistants and the API” is on – the default is off. The lock is checked again before every deleting call. Only things not yet in use can be deleted, such as a question in a draft or a survey without any submission.

How does the assistant tell reading tools from writing ones?

Every tool is labelled as reading or writing as defined in the MCP specification, so claude.ai can allow all reads with one click. Writing tools are additionally marked as additive or overwriting, and invitations as reaching outside the app. Calls are limited per key and minute, and every call is logged with the tool name.

Three example prompts

“Create a customer survey with NPS on three pages, show detractors an extra page with an open question and test every path.”

Claude creates a draft with create_survey_draft, adds pages with add_page and the questions with add_questions, puts the condition “NPS less than or equal to 6” on the extra page with set_page_logic and calls run_auto_test. The reply lists every path with its pages; nothing is published until you ask for it.

“Summarise the open answers of the survey ‘Training 2026’ into themes with quotes and give me the NPS with promoters and detractors.”

Claude follows the summarize_open_answers template: get_survey returns the questions, list_submissions and get_submission_answers the answers, and Claude forms the themes itself. get_nps_analysis returns the NPS. The summary is written by Claude, not by the 121surveys AI feature – so the answers are processed by your assistant’s provider.

“Simulate the share of preference for our Pro plan at 19 euros against the two competitors in the conjoint question, and suggest a two-product portfolio with the highest contribution margin.”

Claude reads attributes and levels with get_conjoint_results, calculates the shares with simulate_conjoint_market and uses optimize_conjoint_portfolio on a portfolio scenario with unit costs to find the best pair. If no scenario exists, Claude creates one with set_portfolio_scenario, from the Analyst role upwards. If the worker is still computing, the status pending comes back and Claude asks again later.

Frequently asked questions

What is MCP?

The Model Context Protocol is an open protocol through which AI assistants call tools of external applications. 121surveys is MCP-first: the MCP server exposes the app’s functions as tools, plus templates (prompts) for typical analyses and resources such as a survey’s results as CSV.

Which assistants does it work with?

Via sign-in (OAuth) with assistants that let you add custom connectors, such as claude.ai or ChatGPT. Via API key with Claude Desktop, Claude Code, OpenAI’s Responses API and your own scripts. Automation tools such as n8n and Make use the REST API with the same key.

Does Claude use the 121surveys AI features?

No. There is no 121surveys AI feature via MCP: categorisation, summaries, AI text analysis and translation only run in the interface. When Claude analyses answers, your assistant’s provider does the processing. The automatic removal of personal data that runs before the 121surveys AI features does not apply to data an assistant reads via MCP.

What does using MCP cost?

Every tool call counts towards your plan’s monthly API quota; this is a protective limit, not billing in credits. Credits are only used by completed and imported responses – test submissions and invitation e-mails are free. Using the assistant itself runs through your own agreement with its provider.

What is not possible via MCP?

Deliberately not available via MCP are the AI text analysis and the other AI features, buying credits, plans and billing. check_survey_translations reads the state of translations – missing fields and unreviewed AI texts – but marking them as reviewed is reserved for a person in the app. Deleting is also impossible without the triple lock.

Who can set up access?

Only people with the Administrator role create API keys; only owners and administrators of the account edit, rotate and revoke other people’s keys. With OAuth, each person connects on their own and gets at most their own permissions. Every role sees and disconnects its own connections under My account → Connected apps; other people’s only owners and administrators of the account, under Manage Workspace → API, MCP & webhooks.

Connect Claude to 121surveys

Create an account, add https://www.121surveys.com/mcp as a connector and let Claude build and test a first draft. Building and testing a questionnaire is free; the developer page lists every tool.

Start for free