Cloud MCP

The Cloud MCP server extends the local MCP surface (decisions, journals, search) across team members. It speaks the MCP protocol over SSE with bearer-token auth and persists shared context to a Postgres-backed pgvector store.

Cloud MCP requires the Team tier or higher. The local MCP server (stdio) is available on every tier including Free.

Endpoint

Deployment	Endpoint
Vupt SaaS	https://mcp.vupt.dev/v1/sse
Self-hosted	https://{your-host}/v1/mcp/sse

Auth

Same Bearer token as the REST API. Token scopes determine which tools the agent can call:

Scope	Allowed tools
mcp:read	search_decisions, list_squads, read_journal
mcp:write	log_decision, write_journal
mcp:admin	manage shared workspaces

Configuring an agent

Most MCP-aware agents take a JSON config with the SSE endpoint:

{
  "mcpServers": {
    "vupt-cloud": {
      "type": "sse",
      "url": "https://mcp.vupt.dev/v1/sse",
      "headers": {
        "Authorization": "Bearer ${VUPT_TOKEN}"
      }
    }
  }
}

Shared decisions

vupt.search_decisions queries the team's shared pgvector store. Per-project ACLs apply: an agent only sees decisions for projects the calling user has access to. The decision payload includes the original prompt context, the chosen path, and the reasoning — so future agents can grep history before re-deciding.

Throughput

The Cloud MCP supports up to 200 concurrent SSE connections per workspace on Team, 1,000 on Business, and is sized per-customer on Enterprise. Connections that idle for >5 minutes are recycled.

Security

All traffic is TLS 1.3. The SSE handler authorizes every event against the user's scopes. A failing authorization closes the SSE stream with event: error and a problem-details payload. No data is shared across workspaces — pgvector queries are scoped to the workspace at the SQL level (WHERE workspace_id = $1), not in application code.