Combined vs MotherDuck: managed Sources or flexible data pipelines?

Compare Combined's managed Sources with MotherDuck Flights, SQL and MCP. Evaluate HubSpot–Stripe answers, sync behavior, access and pricing.

By the Combined team · September 11, 2026

Choose Combined when you want business apps connected, records kept in sync, and an agent answering questions through a managed data layer. Choose MotherDuck when building pipelines, transformations, and data applications is part of the workflow you want. Both support SQL and AI agents; compare how your team wants the records prepared and maintained.

Choose the workflow your team needs

A revenue team asking “Which customers have open deals and recent payments?” needs reliable CRM and billing inputs. A data team building custom enrichment or reusable warehouse models needs room to program that process. Combined manages the path from source connection to query, so your agent can focus on the account question.

Documentation reviewed September 10, 2026 UTC
Your priority	Combined	MotherDuck
Connect business apps for repeated analysis	Managed Sources, dataset selection, sync, grants, and queries in one workflow.	Load through hosted Flights or an ingestion integration, then query the resulting data.
Customize ingestion and transformations	Use the supported connection route and available datasets.	Write or generate Python Flights and maintain reusable SQL objects.
Give an agent business context	Discover actual schemas and supply explicit identity and metric definitions.	Guides store shared definitions and join conventions alongside warehouse tools.
Let an agent build data applications	Supply a bounded, read-only analytical data surface.	Use read/write SQL, Flights, and Dives for pipelines and visualizations.

Compare the Combined MCP interface, MotherDuck's MCP capabilities, and its Guides for business context. Existing data and reusable models change how much work remains before your first useful answer.

MotherDuck can manage the pipeline runtime

MotherDuck Flights execute Python with scheduling, credentials, and run history. Its documentation recommends dlt for ingestion; your team controls the code and dependencies while MotherDuck runs the infrastructure. An agent can create and manage the Flight through MCP. Flights overview.

There are specific ingestion guides for HubSpot and Stripe. Their behavior matters: HubSpot's optional property-history tables have one row per change, while Stripe's example replaces mutable objects and incrementally loads events and balance transactions. MotherDuck also supports ingestion platforms such as Fivetran and Airbyte, giving teams another way to manage connectors. Ingestion options.

Combined brings the Source lifecycle directly into the product. Check current Ready status, authorize the app, select datasets, complete the first successful sync, and grant the agent those Sources. Then discover their actual logical schemas. Your team can focus on the customer map and business calculation while Combined manages ingestion, storage, and query access. Connection steps.

Decide what the agent should be able to do

MotherDuck's hosted MCP connects through OAuth or a bearer token. Its query tool reads data; query_rw can change data or schema. Connection setup, read queries, and write queries.

Roles govern platform capabilities and access to Shares. A Builder role does not automatically expose every Share, and Explorer can create owned resources. For an analysis-only agent, review actual grants and enabled tools, including tools beyond SQL. Read scaling tokens provide a separate route to read-only database replicas, with eventual consistency. Role permissions and read-only options.

Combined uses explicit Source grants and read-only queries. Its MCP returns at most 1,000 rows; the underlying launch policy includes a 256 MiB estimated-input ceiling and 20-second hard deadline. MotherDuck's read-query tool caps output at 2,048 rows or 50,000 characters, with a 55-second timeout. Evaluate your actual query against these bounds, including any truncation. Combined SQL policy and MotherDuck query limits.

Test the answer, then test the next sync

Open the HubSpot–Stripe plan and download the platform evaluation worksheet. Use its blank fields to record your selected sources, setup work, access, freshness requirements, and observed results.

Start with the existing synthetic DuckDB fixture. It combines current open deals with paid invoices whose paid timestamps fall from June 12 inclusive to September 10 exclusive, 2026 UTC. Map scoped Stripe customers to HubSpot companies, then aggregate each fact table separately by company and currency before joining. The fixture converts USD and EUR minor units by dividing by 100.

Expected result from the published synthetic fixture
Company	Open pipeline	Paid invoices in the interval
Atlas	USD 20,000	USD 5,000
Birch	USD 6,000	No match
Cedar	EUR 4,000	EUR 1,500

Birch's older payment and open invoice do not qualify. Missing matches in real data can also mean incomplete syncs or customer mappings. The paid metric includes amounts paid on invoices with paid status; partial payments on still-open invoices are excluded. It is not MRR, all cash receipts in the interval, or revenue after refunds. The full workflow supplies the definitions and expected output.

For the next step, use disposable, authorized source records and choose a reporting interval containing the test activity. Change one deal's value and take an open invoice through payment. Run ingestion again, then repeat the same calculation and time bounds. The changed deal should appear once, the newly qualifying paid invoice should enter once, and unrelated accounts should retain their results.

Record which process handles those updates: your MotherDuck Flight or ingestion partner, or the Combined Source. Check commit freshness and reporting coverage separately. Adapt the teaching aliases to discovered relations before querying either platform. This second-sync exercise is a proposed evaluation, not an observed platform benchmark. Keep the query, record IDs, sync evidence, and mapping exceptions together.

Compare the cost of your actual workflow

MotherDuck documents a $0 Lite platform fee with 10 GB storage and 10 compute unit hours per month included, and Business at $250 per month plus usage. Lite uses Pulse; Business adds other compute sizes and read scaling. Pulse meters resources per query, while larger instances meter elapsed time with cooldown. Compute and storage rates vary by region. Pricing model.

Detailed docs allow Flights on every plan. Cardless Lite permits manual runs within 30 minutes per day and one concurrent run. Adding a card enables scheduling and usage billing; Lite retains a one-hour maximum per run. Current Flight limits. Compare the intended ingestion, storage, queries, and maintenance with Combined's monthly active record model. Include an ingestion partner's cost when that is your chosen route.

Combined's initial allowance is 5 million MAR without a card, then $5 per million MAR. Use the same reporting workload to estimate changed records, query sizes, and ongoing operational work. Neither the Lite platform fee nor the MAR rate alone represents every workload's total cost.

Questions that change the choice

What if our data is already in MotherDuck?

Start with the models and ingestion you already trust. Combined is compelling for a business-app workflow whose authorization, sync, storage, and agent access you want managed together. Compare the remaining work for that specific question, including who maintains its inputs after setup.

Will either product define our metrics automatically?

Your team still needs an agreed customer map, reporting grain, currencies, and time rules. MotherDuck's Guides can retain those definitions for agents; Combined lets the agent inspect granted schemas and apply explicit definitions in SQL. Use the fixture above to check the calculation. For engine and table-format background, see DuckDB and DuckLake for AI agents.

Give your next CRM-and-billing question a managed foundation. Explore Combined, start with the cardless allowance, and follow the self-serve connection steps. Use the plan above to turn two business apps into a repeatable answer.

Sources and further reading

Explore the documentation behind this guide. Product details checked on September 11, 2026.