Skip to main content
This page documents the Python MCP surface (mcp.heybtw.com, Claude Desktop and other MCP clients). It reflects the tool exactly as implemented in mcp-server/tools/workspace_overview.py as of this page’s publication — see Schema stability.

Purpose

get_workspace_overview returns a single-object snapshot of the calling workspace: its identity and plan, which integrations are connected, headline program metrics (signups, MQLs, pipeline), and agent guidance (which other tools are worth calling next, given what’s actually configured).

When to use this

  • The first call in any new agent session against a workspace you haven’t queried before.
  • “What can I ask about this workspace?” or “Is HubSpot connected here?”
  • Before recommending a specific tool to the end user — agent_guidance.data_limitations tells you which tools will return degraded or empty data.

Parameters

Default behavior

Called with no parameters, this tool returns all-time program metrics (program_metrics.period is "all_time") for the calling workspace — no result cap, no pagination, a single object (result_count: 1). configuration.integrations reflects live connection state at call time; it is not cached.

Sample request

Sample response

Response field reference

Envelope

data[0]

data[0].configuration

data[0].program_metrics

program_metrics.attribution_summary — when configuration.integrations.hubspot_crm is false, this shrinks to { crm_connected: false, total_sourced_pipeline: null, total_influenced_pipeline: null, total_closed_won: null, recommendation: "<string>" }. When CRM is connected but the attribution service is unreachable, every dollar field is null and coverage_note explains why (the fuller shape shown in the sample omits coverage_note because it succeeded). avg_cost_per_meeting and avg_cost_per_mql are always null today — cost data is not yet wired into this rollup (see get_event_attribution for the one place program-level cost-per-meeting IS live, under program_summary).

data[0].agent_guidance

Errors

This tool raises no tool-specific structured errors today. Any failure (workspace not found, unexpected exception) is caught by the server’s generic handler and surfaces to the client as:
The error_id is logged server-side with the real exception — useful when filing a support request, but the client never sees which specific condition failed. This is current behavior across all 8 tools; see MCP Tools — Errors and troubleshooting for the transport-level errors (401, 429, 503) that can occur before a tool call is even attempted.

Pagination and limits

Not applicable — this tool always returns exactly one object. There is no limit parameter and no cursor-based pagination on any of the 8 read tools today (see Schema stability).