Hosted MCP Tool Workflow — Advanced Reference

This is an advanced reference for developers, AI operators, and support teams inspecting how an assistant uses Daeda AI. Most RevOps users do not need to call these tools or write SQL themselves.

When You Need This Page

Use it when building an integration, diagnosing incorrect tool use, reviewing multi-portal behaviour, or checking how an assistant prepares a Write Plan. For ordinary setup, follow Start Here and the Integrations guide instead.

Supported Connection

Daeda AI runs as a hosted Streamable HTTP MCP service at https://mcp-ws.daeda.tech/mcp. The discontinued local @daeda/mcp-pro stdio package is not a supported installation path.

Follow the Integrations guide for client-specific connection steps. After the connection succeeds, use the workflow below.

Normal Read Workflow

1. Call get_overview to see connected portals, the selected portal, and supported operation groups.
2. Call set_selected_portal only when the user wants a different default portal.
3. Call get_portal_overview to discover the selected portal’s available tables and operations.
4. Call describe_table before querying an unfamiliar table. Do not guess column names.
5. Call query with read-only SQL. Use explicit portalIds for intentional multi-portal work.
6. Use refresh_data when selected synced data needs a manual refresh. Use fetch_on_demand only after identifying exact record or list IDs.

status(section="schema") is a broad diagnostic snapshot. It is not the normal starting point for portal discovery or query planning.

Normal Write-Plan Workflow

1. Use get_overview or get_portal_overview to find the exact supported operation types for the target portal.
2. Call describe_operations for only the operation types needed for the requested change.
3. Call build_plan to create or revise an editable draft. Keep dependent operations together and use plan references instead of guessing IDs.
4. Call submit_plan with dry_run: true to validate while preserving the editable draft.
5. Present the validated draft and wait for the user’s approval before saving it.
6. Call submit_plan without dry run to save the plan as DRAFT for review in Daeda AI.
7. Use get_plan or get_plans to inspect saved state. Execution still happens through the human review surface in HubSpot.

Hosted MCP Tool Reference

Tool	Use it for
build_plan	Create or revise the editable write-plan draft.
describe_operations	Return the exact contract for selected write operation types.
describe_table	Return exact table columns and fill rates before a query.
fetch_on_demand	Fetch scoped property history or list memberships after IDs are known.
get_overview	List connected portals, the selected portal, and supported operation groups.
get_plan	Open one saved plan and its current results.
get_plans	List saved plans by portal and optional status.
get_portal_overview	Show data, tables, and operation availability for a portal.
query	Run read-only SQL against authorized mirrored HubSpot data.
refresh_data	Queue a refresh for selected mirrored data or metadata.
set_selected_portal	Select the default HubSpot portal for later calls.
status	Inspect broad schema or metadata diagnostics when troubleshooting.
submit_plan	Validate the draft or save it as a reviewable DRAFT plan.

The old local-only chart and configure_sync_controls tools are not exposed by the hosted MCP service. Use refresh_data for refreshes and set_selected_portal for portal selection.

Manual Verification Prompt

Ask your connected assistant:

Use Daeda AI to show my connected portals, tell me which portal is selected, describe the data available for it, and do not make any changes.

A healthy answer names the portals, selected portal, and available data without using any write tool.