Composio review: tools, account scope and repeatable business-data answers
A practical Composio review covering sessions, direct tools, account scope, exports and cost units. See when to choose Combined for managed business-data SQL.
For recurring SQL questions across business apps, we recommend Combined: its managed ingestion, stored datasets and read-only MCP query path give the answer a maintained data foundation. Composio is a strong option for agents that discover and execute app tools with managed authentication. This review examines the operational work behind those choices, including when to use both.
Our verdict: choose the platform around the deliverable
Choose Combined when your main deliverable is a repeatable business-data answer. CRM pipeline, paid invoices and account activity need complete inputs, explicit joins and stable definitions. Combined syncs selected records into managed storage and exposes them through SQL and MCP. Your agent can discover the schema, inspect freshness and rerun the calculation.
Composio deserves consideration when the agent's job centers on app operations. Its sessions combine a user, account access, tool discovery and execution state. They support broad agents that choose tools at runtime, as well as narrower configurations. You can use Composio for approved actions alongside Combined for the data analysis.
This is a Combined-authored architecture review of public documentation checked on September 17, 2026. The recommendations are based on the work each approach requires; this is not a live vendor benchmark. For a wider shortlist, use Composio alternatives.
Three execution choices that should not be conflated
| Mode | Tool selection | What to evaluate |
|---|---|---|
| Session with discovery | Meta tools discover and execute operations during the task. | Allowed toolkits, accounts, dependencies and behavior when another connection is required. |
| Session with direct tools | The preset exposes the permitted app tools upfront; meta tools are disabled by default. | The exact tool list and any connection-management or sandbox helpers you re-enable. |
| Direct execution | Your application chooses schemas and calls tools outside a session. | Your auth flow, state, toolkit version and orchestration. |
These distinctions come from Composio's execution guide and configuration reference. A direct-tools session still has session identity and account configuration. It is different from calling the standalone tool execution API.
For a narrow recurring task, inspect the actual exposed tool list and retain the configuration with your release. For an exploratory assistant, evaluate discovery and the fallback when a necessary tool is absent. Composio documents version pinning for direct execution; sessions manage toolkit versions automatically. Include a regression check for the fields your report depends on when changing your SDK or execution path.
Review account selection before reviewing the answer
A stable application user ID gives Composio an identity under which to keep connections. A user may have more than one account for a provider. The account-selection rules support an explicit connected-account override and describe fallback behavior that can select the most recently connected account. For a customer report, pin the intended account rather than relying on recency.
- Record the application user, provider account and required permissions for the workload.
- Limit enabled tools; inspect read operations and any broader proxy or sandbox execution path.
- Test a second account at the same provider and confirm that the report still uses the intended one.
- Check the behavior after a connection becomes inactive or requires renewed authorization.
Composio's account lifecycle covers credential refresh and connection state. Connected-account responses describe authentication; sensitive credentials are redacted by default. Listing those accounts does not retrieve the customer's CRM records or create a portable OAuth grant.
In Combined, configure the target Account and explicit Source grants for the agent's identity. Use dataset allowlists and applicable query bounds for the report. The direct comparison explains how the two access paths fit their different workloads.
Composio can retrieve and process data; define the complete report path
The HubSpot toolkit includes tools such as HUBSPOT_LIST_CONTACTS and HUBSPOT_LIST_DEALS. Inspect the schema for the selected version, including which properties and pagination inputs it accepts. The Tools reference also documents authenticated proxy requests for provider endpoints; Composio adds the credential server-side.
Its remote sandbox can process bulk results and keep working state within a session. The persistent /mnt/files/ mount survives a sandbox restart; its experimental files API can transfer files between the application and that mount. That is a real persistence capability to include in the assessment.
For an ongoing business report, define the dataset lifecycle you intend to operate: initial coverage, subsequent updates, deletions, page completion, failure recovery and the point at which an output is ready. Decide where the accepted export lives and who refreshes it. A useful account lookup can succeed before those report requirements are met.
Follow our Composio provider-data export guide for a bounded retrieval example and its validation scope. If the requirement is a maintained queryable dataset, start with Combined's managed ingestion-to-SQL path. You still define the business logic, but you do not need to assemble a separate storage and MCP query service for that path.
Price the execution route you will actually use
As checked September 17, Composio advertises 100,000 monthly tool calls free for own-app/API-key/MCP usage. Scale is $29 monthly with $29 usage credit; its base overage is $0.0003 per tool call. Search meta tools are free. Managed OAuth apps have lower allowances and different rates. See the current rate card and your account's terms.
| Record | Include in the estimate |
|---|---|
| Credential route | Own app versus Composio-managed app; applicable allowance and connection charges. |
| Execution | Tool calls, delivered triggers and applicable direct, proxy or sandbox add-ons. |
| Other usage | Premium provider tools, optional sandbox LLM tokens and enabled account add-ons. |
| Application operation | Your model, retained outputs, reporting runtime and maintenance work. |
The rate card lists a sandbox-execution surcharge and separate sandbox LLM usage. The sandbox guide says compute-tier billing is still planned. Treat compute tiers, tool executions and LLM tokens as separate pricing questions; confirm all three for your configuration.
Combined charges by changed records at its trusted commit boundary: $5 per million MAR after the trial, prorated, with an initial cardless 5 million MAR. A tool call and a record measure different work. Use the evaluation worksheet to estimate the same workload and ownership for each approach, then check Combined pricing and query limits.
Use one complete answer to make the decision
Evaluate a real question such as open CRM pipeline alongside paid invoice amounts. The HubSpot–Stripe tutorial supplies a runnable synthetic SQL fixture and expected output. Use those values as a calculation check, then adapt the logic to authorized data.
- Coverage: confirm every required object and field, complete pagination and a defined reporting interval. Keep incomplete retrieval distinguishable from an empty dataset.
- Mapping: maintain the CRM company-to-billing customer map; review missing and ambiguous links. The provider IDs are different identities.
- Arithmetic: aggregate deals and invoices separately before joining by company and currency. Paid invoice amounts do not automatically represent MRR or revenue net of refunds.
- Rerun: change an authorized test record, repeat the report and inspect its data age. Record who handles stale data, a changed schema and failed access.
Use the freshness checker to set acceptable ages before the evaluation. Keep the observed result, configuration and operating work beside the monthly usage estimate so your team can make the decision from the same evidence.
Where we recommend Combined, Composio or both
Choose Combined for recurring business-data analysis. It brings managed ingestion, queryable records, schema discovery, freshness and scoped SQL access into one data path. Start with the business-data planner, then connect your selected Source and agent. For a client project, use our FDE delivery plan.
Choose Composio when managed authorization and app-tool execution are the center of the product you are building. Use both when an approved action follows a business-data calculation: Combined supplies the analysis; your action integration receives the reviewed inputs and executes the approved operation.
Comparing integration infrastructure first? Read Composio vs Nango for current session, discovery and sync capabilities, or the alternative shortlist. Ready to build the data path? Start with Combined and follow the Claude Code setup guide.
Sources and further reading
Explore the documentation behind this guide. Product details checked on September 17, 2026.
- Combined MCP tools and Source grants
- Combined SQL interface
- Combined connector availability
- Combined pricing and MAR
- Composio sessions
- Composio configuration and account selection
- Composio execution modes
- Composio account lifecycle
- Composio toolkit versions
- Composio HubSpot tools
- Composio tool execution and proxy
- Composio sandbox and persistent files
- Composio pricing