Connect other MCP clients
Avlok’s built-in MCP connector is built on standard MCP, so it is not limited to Claude and ChatGPT. Any client or agent that speaks the Model Context Protocol can connect to your photo library. For an overview of the connector and the verified assistants, see The Avlok MCP connector.
Two endpoints
Section titled “Two endpoints”Local, over stdio. The bundled avlok_mcp binary runs on your machine and speaks MCP over standard input and output, the same transport Claude Desktop uses. Any client that can launch a local stdio MCP server can use it. Point your client at the binary; the path is filled into the configuration snippet in Avlok’s Settings → AI Integration.
Cloud, over Streamable HTTP. With Cloud AI Access on, Avlok serves your chosen folders from a hosted endpoint over MCP’s Streamable HTTP transport, authenticated with OAuth. Any client that supports a remote MCP server with OAuth can connect. Clients that cannot run an OAuth flow can use a token generated in Avlok’s Cloud AI Access settings instead; see Authentication. The connector URL is in Settings → AI Integration.
What a client gets
Section titled “What a client gets”Either endpoint exposes the same shape of capability: a set of tools for searching and describing your library. The local endpoint also exposes tools that depend on desktop-only features, such as searching the words spoken in your videos. See the Tool inventory for the full list and which surface each is available on.
Protocol details
Section titled “Protocol details”For the version of MCP supported, the transports, and the authentication flow, see:
A note on building your own
Section titled “A note on building your own”If you are building a client or an agent against Avlok, the tool reference documents each tool’s inputs and outputs. Tools are read-only: they find and describe, they never modify your library. Treat the cloud endpoint’s results as scoped to the folders the user has chosen to share.