Claude Code can do more with company context than another spreadsheet export. Connect it to the business records it needs, give it a clear question, and let it query those records through MCP.
We build Combined, a managed data layer for this workflow. Combined connects to business applications, keeps selected data synced, and exposes scoped, read-only access to compatible agents. Here is the setup path for Claude Code.
Prepare one useful dataset
Start with one source and a question you already care about: open pipeline by stage, for example. In Combined, choose a Ready connector, complete authorization, select the relevant datasets and fields, and run the first sync. Check the records in Explore.
Give the agent access to those datasets through the onboarding or Access workflow. The scope should match the question. You can expand it when a later question needs another dataset.
Connect Claude Code
Use the account-specific instructions supplied by Combined. Replace ACCOUNT_UUID below with your account ID, then add the HTTP server in Claude Code:
claude mcp add --transport http combined-context \
"https://platform.trycombined.com/mcp?account_id=ACCOUNT_UUID"
Open /mcp in Claude Code and follow authentication. The account ID selects the account; authentication provides access. Keep credentials in the supported setup flow.
Ask the agent to discover the data
Business systems differ, so let the agent inspect the available source, logical dataset, and field names before writing SQL. A useful first instruction is:
Use the combined-context MCP server.
List my permitted sources and use the one I select.
List its datasets and describe the dataset needed for my question.
Check freshness, then run a read-only query returning at most five rows.
Report the source, freshness, query receipt ID, and any truncation.
Once that works, ask a precise business question. For pipeline, specify the pipeline, time boundary, stages, and currency treatment. Have the agent aggregate the relevant records rather than infer a total from a small row sample.
Make the answer useful
The best answer gives you the result and enough context to act on it: which source it used, when that data last synced, and any missing coverage. If a record just changed in the source, sync the change before expecting it in the result.
Start with one question you would otherwise answer manually. Connect its source once, then keep the question inside your existing agent workflow.
Combined includes an initial 5 million MAR without a card. Follow the full connection guide or see the HubSpot pipeline example.
References: Combined MCP documentation and Claude Code MCP documentation.
Top comments (0)