Direct answer: Finta MCP uses OAuth for end-user connections, binds access to an authorized Finta organization, derives user identity from the validated token, and requires a second confirmation for selected consequential actions. Security also depends on the MCP client, the permissions you grant, and the accuracy of the underlying data.
Security model at a glance
| Layer | Finta control | User or client control |
|---|---|---|
| Authentication | OAuth bearer validation for end-user sessions | Sign in only through the expected Finta authorization screen. |
| Organization scope | The connection binds to an eligible organization membership | Select the intended organization and disconnect stale access. |
| User identity | Identity is taken from the validated token, not a tool argument | Do not share sessions or approve unknown clients. |
| Tool access | Finta publishes a defined tool inventory and applies permissions, rate limits, and credits | Disable tools that are not needed in the client. |
| Consequential actions | Selected tools return a proposal and short-lived confirmation token before execution | Review recipients, record IDs, dates, amounts, and blast radius. |
| External apps | Each app connector has its own scopes and provider controls | Grant the smallest provider scope and revoke unused connections. |
Before connecting an MCP client
- Confirm the endpoint is exactly
https://finta.ai/mcp. - Confirm the browser is on a Finta-owned domain before signing in.
- Use an individual account, not shared credentials.
- Choose the correct organization during OAuth.
- Read the client provider's data and retention terms.
- Keep tool approval enabled while testing.
- Avoid broad research with write-capable tools enabled.
Understand confirmation tokens
For selected high-impact tools, the first call does not execute the action. It returns a human-readable summary, the affected recipients or records, and a short-lived confirmation token. Execution requires a second call with the same arguments and that token.
This pattern is used for actions such as email sending, calendar creation, selected automation changes, contact enrollment, and destructive stage or automation operations. It reduces accidental execution, but it does not remove the need for human judgment.
Protect relationship context
Relationship intelligence can include private notes, communication history, preferences, timing, and network context. Before exposing it to a client:
- Confirm the client and workspace are approved for that data.
- Limit the task to the smallest necessary set of records.
- Avoid placing sensitive data in reusable prompts or shared projects.
- Review any output before forwarding it outside the organization.
- Disconnect the client when the project ends or a user's role changes.
How Finta helps
Finta keeps the durable CRM and relationship record inside the authorized organization. MCP provides a controlled interface to supported tools, not an unrestricted database export. Finta applies its own authorization and action rules even when the client also has approval controls.
Limitations and disclosure
- No system can guarantee that a connected model will always be correct or resistant to every prompt injection.
- Third-party clients and app connectors have separate terms, retention policies, and security controls.
- Finta MCP is not a substitute for legal, regulatory, recordkeeping, or security review.
- An authorized user can still approve an incorrect action. Review is an operational requirement.
Related guides
Last verified: August 21, 2026.