Browse documentation

MCP overview

Preview This capability is in Preview and may not be enabled for your organization yet. About release stages

Ask your AI assistant questions that are answered directly from your published Neuralift segments.

Neuralift includes an MCP server: a secure endpoint that AI assistants such as Claude, ChatGPT, and Microsoft Copilot can connect to using the Model Context Protocol. Once connected, your assistant answers questions from your real published segmentation results instead of guessing from general knowledge: segments, personas, insights, recommended actions, action plans, segment tactics, and the data dictionary behind them.

That turns everyday questions into grounded answers:

  • “Which segment has the highest expected-impact segment tactics, and what’s the evidence?”
  • “Describe the persona for segment 7 in language I can share with the creative team.”
  • “Draft a campaign brief for my highest-value segment using its persona and the action plans that target it.”
  • “What does the days_since_last_purchase column mean, and what are typical values?”

How it works

Your assistant connects over HTTPS with an organization-scoped API key. Every request is read-only and scoped by that key.

How an assistant connects to the Neuralift MCP server Your AI assistant Claude, ChatGPT, Copilot Studio HTTPS + API key published results only Neuralift MCP server read-only, org-scoped API key
Requests go in with an API key; only published results come back.

The endpoint is https://app.neuralift.ai/mcp. Your Neuralift team will confirm your exact endpoint when enabling MCP.

The trust boundary

The MCP server is designed so that connecting an assistant cannot change anything in Neuralift or expose work in progress:

BoundaryWhat it means
Read-onlyAll nine tools are read operations. There is no way to create, edit, or delete anything through MCP.
Published runs onlyRuns, segments, insights, personas, segment tactics, actions, and action plans are returned only from use cases that Neuralift has published; the results of draft and in-progress runs are not returned, even to keys created by administrators. The one exception is the data dictionary: prepared-dataset and column metadata (names, descriptions, sample values, statistics) is visible to any valid key for your organization regardless of publication state.
Scoped to your organizationEvery request is authorized by an API key that belongs to one organization. The key, not the person who created it, determines what data is visible, and it can never see another organization’s results.
No PIINeuralift does not accept any data with PII, so segment results contain none.

Where your results travel

Note. Connecting an assistant means the data these tools return (segment profiles, narratives, business context, and dictionary metadata) is sent to that assistant’s provider and handled under their terms. Review your assistant provider’s data-retention and training policies before connecting, and use a dedicated API key per integration so you can revoke each one independently.

Enabling MCP

MCP is available by request and under agreement only; it isn’t switched on with your standard plan. Contact your Neuralift team or support@neuralift.ai to request access; once the agreement covering MCP access is in place, Neuralift enables the capability for your organization and confirms your exact endpoint. Once enabled, organization admins see an API keys tab in Settings where they can create keys (see roles and permissions).

Connect your assistant

AssistantSupportGuide
Claude CodeVerified pathConnect Claude
Claude Desktop and claude.aiDocumented; static-key authentication pending verificationConnect Claude
OpenAI (Codex CLI, ChatGPT)Codex CLI supports API-key authentication directly; ChatGPT is documented, pending verification of static-key authenticationConnect OpenAI
Microsoft Copilot StudioDocumented; static-key authentication pending verificationConnect Microsoft Copilot

Each guide has the details, including any authentication caveats for that client.

When you’re connected, browse the tools reference to see everything your assistant can look up, and borrow from prompts and workflows to get useful answers quickly.