ドキュメント
MCP
Connect ChatGPT, Claude, Cursor, or any remote MCP client to Schift with your Schift API key.
Use Schift's hosted MCP endpoint when you want an AI client to search your Schift memory, buckets, and approved workflow tools without running your own MCP server.
Connect the hosted server
| Field | Value |
|---|---|
| Server URL | https://mcp.schift.io/mcp |
| Authorization | Bearer <your-schift-api-key> |
Use your Schift key
The hosted MCP server passes your bearer token to https://api.schift.io, so every tool call runs with your Schift permissions. Do not use a separate MCP bearer token for mcp.schift.io.
OpenAI Responses API example
json{
"type": "mcp",
"server_label": "schift",
"server_url": "https://mcp.schift.io/mcp",
"headers": {
"Authorization": "Bearer <your-schift-api-key>"
},
"allowed_tools": [
"search",
"fetch",
"schift_search",
"schift_upload_document",
"schift_memory_search",
"schift_workflow_list",
"schift_workflow_dry_run",
"schift_workflow_run"
],
"require_approval": "never"
}What the tools do
| Tool | Use it for |
|---|---|
search / fetch | Search inside Schift buckets and fetch cached result text. This is not external web search. |
schift_search | Search inside a specific Schift bucket or collection |
schift_upload_document | Upload text or base64 file content into a Schift bucket and queue async indexing |
schift_memory_search | Search the authenticated user's connected memory sources |
schift_workflow_list | List workflow tools the user can see |
schift_workflow_dry_run | Preview a workflow run before external effects |
schift_workflow_run | Run a published workflow, with Schift review boundaries still enforced |
Self-hosting is optional
Only run schift-mcp yourself when you need a private MCP deployment. For normal hosted use, connect directly to mcp.schift.io with your Schift API key.