Outset

What the Outset MCP server does

The Outset MCP server exposes Outset’s research workflows to AI agents through the Model Context Protocol. It is intended for teams using Outset to plan studies, recruit participants, and review research output through an MCP-compatible client.

Agents can create and modify study components such as projects, sections, questions, screeners, display logic, quotas, and concepts. They can also configure recruitment with a sample size, reward, and demographic filters. Recruitment launch is available only when the authenticated account has the separate studies:launch scope, and launching may incur fees.

After research is conducted, the server provides tools for querying reports, answers, topline insights, category summaries, and highlight clips. Other tools expose workspace, project, and organization information so an agent can work with the relevant Outset context.

How it works

The Outset MCP server is hosted by Outset at https://api.outset.ai/mcp/. It uses stateless JSON-RPC over Streamable HTTP rather than a local process. A client connects to the endpoint and completes an OAuth 2.1 flow using PKCE. Dynamic Client Registration means compliant clients do not need to be registered manually before connecting.

Permissions and organization roles are checked by Outset. Available tools can change as new capabilities are released, so the live tools/list response is the authoritative tool inventory. The server’s supported scopes and role requirements are documented in Outset’s separate developer integration guide.

Setup and configuration

No package installation or local runtime is required for the Outset MCP server. Add the endpoint to an MCP client that supports remote servers over Streamable HTTP and OAuth, then sign in with an Outset account.

Claude, Cursor, ChatGPT, VS Code, Gemini CLI, and Codex each provide connection methods documented by the project. Claude Code can add the server with its HTTP transport command or install the repository’s plugin. For clients that support only stdio, mcp-remote can bridge the remote endpoint:

npx -y mcp-remote https://api.outset.ai/mcp/

The repository contains client and directory manifests, including configurations for Claude Code, Cursor, Gemini CLI, and the official MCP registry. It does not contain the hosted server implementation.

Tools and capabilities

Supported operations include:

• Creating and editing studies, sections, questions, screeners, display logic, quotas, and concepts.
• Setting recruitment sample sizes, rewards, and demographic filters.
• Launching recruitment when the account has the required scope.
• Reading reports, participant answers, topline insights, category summaries, and highlight clips.
• Looking up workspaces, projects, and organization information.

The exact tool names and current inventory should be obtained from the endpoint with tools/list.

Limitations and notes

The Outset MCP server requires an Outset account and an MCP client capable of the required remote transport and authentication, unless a stdio bridge is used. Client-specific subscription requirements may apply; for example, custom connectors in Claude require a Pro, Max, Team, or Enterprise plan.

Access is not uniform across accounts. Server-side scopes and organization roles control which operations are available, and launching studies can require the additional studies:launch scope and may generate fees. Because the tool set changes over time, integrations should rely on the live discovery response rather than assuming a fixed list.