The Model Context Protocol (MCP) lets other AI clients call tools from Corbis.
That means you can use Corbis research and data tools inside products like Cursor, Claude Code, ChatGPT, and other MCP-compatible clients instead of staying only inside the Corbis web app.
MCP is useful when you want to:
The exact tool list can change over time, and availability depends on your account and MCP key settings.
In practice, Corbis currently exposes tools in categories like:
If you only need one mental model, use this one: MCP gives another AI client access to the same kinds of research actions Corbis can perform directly.
For most users, the simplest setup is a personal MCP API key from Corbis.
In Corbis, open Settings → API Keys, create an MCP key, and copy it when it is shown.
Important:
Use your Corbis origin (the same host you use in the browser for the app—not a marketing URL unless that is where the app actually runs). The MCP path you need is:
/api/mcp/universal
Build the full URL as: the same origin you use for the web app (scheme, host, and port if any) plus the path /api/mcp/universal.
Most clients connect with that URL plus your MCP key.
Corbis already helps with the common cases:
If you are not sure which transport to use, prefer the streamable HTTP setup against /api/mcp/universal unless your client docs require something else.
Personal MCP API keys — Created in Settings → API Keys. This is the default path for editors and local MCP clients (Cursor, Claude Code, and similar).
OAuth — Used when a supported integration (for example a hosted assistant product) implements Corbis’s OAuth flow so users sign in and approve access in the web app. You do not need OAuth for a simple “URL + MCP key” setup.
If the client accepts it, send your key as an Authorization: Bearer … header. OAuth access tokens use the same header style after you complete the integration’s login flow.
Security: Some clients allow putting the key on the URL as a query parameter. That can be easier to leak (logs, referrers, shared links). When the client supports header-based auth, prefer Bearer tokens or OAuth over query-string tokens, especially in untrusted or shared environments.
OAuth scope and consent details for implementers are documented in lib/mcp/OAUTH2_IMPLEMENTATION.md in the repository (not required reading for basic API-key setup).
Prefer /api/mcp/universal for new setups. Other routes exist for compatibility and discovery:
/api/mcp/sse and /api/mcp/message — Legacy SSE transport pair for clients that need a dedicated SSE URL instead of streamable HTTP on the universal route./.well-known/oauth-protected-resource, /.well-known/oauth-authorization-server, /.well-known/openid-configuration — OAuth / OpenID discovery metadata for clients that probe well-known URLs./api/mcp/oauth/register, /api/mcp/oauth/authorize, /api/mcp/oauth/token — Dynamic registration, authorization, and token exchange for supported OAuth integrations.This section names routes only; it is not a full OAuth tutorial.
/api/mcp/universal unless you know you need the legacy path.Tool access depends on your account entitlements and key configuration. The safest answer is to check your current plan and limits in Corbis.
Yes. That is one of the main MCP use cases.
Yes. Many users create separate keys for different clients so they can revoke them independently.
Start in Settings → API Keys and make sure you are using the right endpoint and token. If that still fails, contact support.
If you are working from a checkout of the Corbis codebase:
docs/ai-sdk/mcp/ — Corbis MCP guides (Cursor plugin, Claude Code, Codex, and related).lib/mcp/CHATGPT_GUIDE.md — ChatGPT-oriented setup notes.