MCP endpoint
Let your AI agents call Spooling Cloud as a tool via the Model Context Protocol.
Spooling Cloud exposes a Model Context Protocol endpoint so any MCP-aware agent (Claude Code, Cursor, Codex, your own build) can search and read your team's session memory as a tool.
Endpoint
https://api.spooling.ai/api/v1/spooling/mcpProtocol: HTTP-streamable, version 2025-03-26.
Auth: Authorization: Bearer sk_<token> with kb:read (or admin) scope. A key can only see the workspace it was minted for.
Tools exposed
| Tool | What it does |
|---|---|
search_knowledge | Search the workspace's sessions and messages for a query. |
list_sessions | Newest sessions, optionally filtered. |
get_session | A session and its ordered messages, by ID. |
get_stats | Counts, cost and a per-provider rollup for the workspace. |
list_workspaces | The workspaces the key's owner belongs to. |
Local MCP server
Running spooling mcp launches the MCP server on http://127.0.0.1:3004/mcp. By default it queries the local DB and falls back to the cloud API when local data is empty. Control this with mode flags:
| Flag | Behavior |
|---|---|
--hybrid (default) | Local DB first, cloud fallback when empty |
--local | Local DB only, never calls cloud API |
--cloud | Cloud only, skips local DB entirely |
spooling mcp --cloud # cloud-only mode
spooling mcp --local # local-only mode
spooling mcp --hybrid # explicit hybrid (default)Wire it up in Claude Code / Cursor / Codex
Drop this into your client's mcp.json (Claude Code: ~/.claude/mcp.json, Cursor: ~/.cursor/mcp.json, etc.):
{
"mcpServers": {
"spooling": {
"type": "http",
"url": "https://api.spooling.ai/api/v1/spooling/mcp",
"headers": {
"Authorization": "Bearer sk_<token>"
}
}
}
}Replace sk_<token> with a key you minted in the workspace whose memory you want the agent to call.
Smoke test
curl -s https://api.spooling.ai/api/v1/spooling/mcp \
-H "Authorization: Bearer $SPOOLING_KEY" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | jqYou should see the five tools above.
Why this matters
Agents that call search_knowledge before proposing a fix can lean on what the team has already figured out. Agents that don't are starting from zero on every task. The same backing data powers the web app search and the REST API; the MCP endpoint is just the agent-shaped door into it.