What is Combined? Business data for the AI agents you already use
How Combined connects SaaS data, keeps it queryable, and gives external AI agents scoped access through SQL, MCP and SDKs.
Combined is a managed data platform for AI agents. It connects business systems, maintains queryable data, and lets your own agents read permitted datasets through SQL, MCP or typed SDKs. You bring the agent; Combined supplies the data and access controls.
What does Combined handle?
An agent answering a business question needs more than an API token. It needs to know which datasets exist, what their fields mean, when they last updated, and whether its identity may read them. Combined brings ingestion, managed storage and agent access into one account boundary.
- Connect: authorize a supported source, discover datasets and fields, and select what to ingest.
- Keep current: run a manual sync or choose an available daily or hourly schedule. Freshness reflects committed data; schedules depend on the source and account policy.
- Grant: give a person or agent explicit access to selected data with applicable query limits.
- Query: discover logical relations and run bounded, read-only SQL through an interface your agent supports.
- Inspect: use query receipts, source state and freshness to understand what the answer used.
The product documentation describes this as a context layer for external agents. Combined does not build or host the agents themselves.
Who is it for?
Combined is designed for developers, forward deployed engineers and teams giving agents access to business data. A useful first project is a narrow question over one or two supported systems, such as summarizing a permitted dataset or calculating a metric from structured records.
| Your requirement | What to evaluate |
|---|---|
| Read and join structured business records | Combined’s dataset discovery, SQL support and source coverage. |
| Use a coding agent with business context | The account-scoped MCP endpoint and your client’s authentication support. |
| Create tickets or change CRM records | A separate write integration. Combined’s agent data surface is read-only. |
| Find passages by semantic similarity | A semantic retrieval system. Combined’s search_context tool performs literal substring search. |
| Use a particular SaaS connector immediately | Its live Ready status, supported datasets and authorization requirements. |
A catalog entry alone is insufficient evidence that a connector is ready for production. Check its status before designing a workflow around it. Preview access can require a request; it should not be treated as an available connection.
What counts as a successful first connection?
Start by creating a workspace and choosing a Ready connector. Authorize it with the appropriate account, select the datasets and fields you need, and wait for the first successful sync. A completed authorization screen is only one step: verify that the expected records actually appear.
Next, grant yourself the selected datasets and run a preview or SQL query. In the onboarding flow, use Use it with AI to copy the agent setup instructions. The useful finish line is an agent identity producing a successful query receipt over real, permitted data.
If a source is waiting for somebody else’s authorization, you can connect a different source while that request remains pending. If the first sync fails, follow the source recovery steps instead of making duplicate connections.
Follow account onboarding for the complete flow, or use the Claude Code tutorial for an MCP client example.
What does Combined cost?
The public launch price is $5 per million monthly active records (MAR) after the trial, prorated, with no public base charge or minimum. You can start with 5 million MAR without a card. Adding a card unlocks another 20 million free MAR and starts a seven-day card-backed trial.
For example, 20 million billable MAR at the public rate is $100. That is a usage example, not a required $100 monthly subscription. MAR measures changed records at the trusted commit boundary; a source’s raw row total is not automatically its billed usage.
Query limits also apply. The documented defaults include a 100 GiB monthly estimated query input budget per account and a 256 MiB estimated-input ceiling per query. Account-specific commercial terms and stricter grants can change the effective limits. See pricing, examples and trial terms before connecting a large dataset.
Sources and further reading
Explore the documentation behind this guide. Product details checked on September 9, 2026.