Sprig MCP is designed so an AI client sees only the survey data your Sprig role already permits, scoped by OAuth and enforced at every tool call. Agents can create studies in draft only, a human approves before anything reaches respondents, and admins hold an org-wide kill switch. Sprig never trains on your responses, and unlike a CSV export, MCP leaves no permanent, unscoped copy of your data.
Data Handling and Retention
1. What happens to Sprig data inside the LLM conversation? Is it retained or used for training?
When your team prompts Sprig through MCP, the returned data becomes part of that conversation on the LLM provider's side. MCP introduces no new data exposure category beyond what your organization already accepted by adopting the LLM tool. Sprig does not store copies with the provider and does not control the provider's retention.
- In transit, tool calls run over TLS between the client and mcp.sprig.com.
- In context, the data lives inside the conversation for its duration.
- After the conversation, retention, training treatment, and provider-side access are governed by your organization's agreement with the LLM provider, not by Sprig.
2. Does Sprig or the LLM provider train models on our response data?
Sprig does not use your response data to train models, full stop. Data returned through MCP is then subject to your LLM provider's own data handling terms, so enterprise customers should confirm their Claude, OpenAI, Gemini, or Copilot agreement excludes training on API and connector traffic. Enterprise plans from Anthropic, OpenAI, and Microsoft exclude that traffic by default, and we're glad to point your security team to the relevant clauses.
3. How is MCP different from exporting a CSV and uploading it to an LLM?
A CSV export is a permanent, unscoped copy; MCP is scoped, live access with no lasting copy. Once a CSV is downloaded, it sits outside Sprig's permission model: anyone with the file can share it, upload it anywhere, and keep it indefinitely, with no way for you to revoke it. MCP inverts that. Data is fetched at prompt time, scoped to what the authenticated user is allowed to see, and access ends the moment an admin revokes the token or disables the server. A CSV upload is a one-time transfer with zero ongoing control; MCP is live access with your controls intact.
Access and Permissions
4. What data can an LLM client access through Sprig MCP?
An LLM client can reach the same research data the authenticated user can already see in Sprig: survey definitions, response data, open-text themes, and response counts. It pulls this at prompt time and only within that user's role scope, never as a bulk export or a standing copy of your workspace. Responses are returned in capped batches (up to 1,000 per call) rather than a single bulk dump, and any personal data in responses follows the same field-level controls it does in the Sprig app.
5. How is access authenticated and controlled?
Every Sprig MCP connection runs through OAuth. Users authenticate with their own Sprig credentials, and the MCP server enforces the same role-based permissions as the Sprig app. A user cannot see data through Claude, ChatGPT, or Cursor that they could not see logged into Sprig directly. Because authorization is tied to the individual user's role, access reflects your existing Sprig permission model rather than a separate one.
6. How do LLM clients handle Sprig permissions on their side?
The LLM client never holds standing access to your Sprig workspace. It holds an OAuth token tied to one user, and every tool call is authorized against that user's Sprig permissions at request time. Change a user's role or revoke their token and the client's access changes with it. Admins on enterprise LLM plans also control which connectors are available to their org at all, adding a second permission layer on top of Sprig's.
Agent Actions and Control
7. Can an AI agent launch or modify live studies through MCP?
No. Sprig MCP exposes study creation in draft state only. A human reviews and approves every study in the Sprig app before it reaches respondents. MCP-created drafts are indicated as such, so your team can identify their origin during review. An agent cannot push a study live, change a running study, or reach respondents on its own.
8. What happens if we need to shut off access immediately?
Admins can turn the MCP server off from Integrations, AI & MCP in Sprig. This kill switch revokes all active connections org-wide, immediately, without waiting on individual users to disconnect their clients. It gives security and IT teams a single control to cut off all MCP access at once.
Connect your workspace to Sprig MCP at mcp.sprig.com/mcp.
%20(1).png)
