Connect local applications and Agents
Installation, sign-in, authorization and execution are separate checks.
Check available executors
Install the required application or Agent on the target computer and verify that its account works. Ecosystem artwork does not imply that every product has the same integration or permission model.
Authorize for the task
Select only the directories, apps and tools the project needs. OS permissions, third-party account access and project grants are all required. A read-only research task should not receive unnecessary write or send access.
Run a small verification task
Use a non-sensitive test file for a small read, generate or review task. Inspect the actual result and run record before expanding scope. Successful configuration is not proof of successful execution.
Handle lost access
Expired accounts, app updates, moved directories and revoked OS permissions can interrupt work. Check the failure, then reauthorize or revise the task. Avoid granting whole-device access just to bypass a failure.
Choose an integration route
An app may be connected through its official API, an MCP server or UI operation. Assess stable interfaces first; use UI operation only where authorization, application terms and runtime conditions allow it. Validate read and write scope for each route.
| Route | Prerequisites | Maintenance |
|---|---|---|
| API / connector | Account, API permissions and test environment | Token renewal, rate limits and schema changes |
| MCP | Trusted server, tool catalog, configuration and credentials | Protocol compatibility, side effects and health |
| UI operation | Signed-in app, OS permissions and stable session | UI changes, session expiry and result checks |
Keep an integration register
Record the owner, account, scopes, connection route, version, last verification and revocation steps. Disable access when a person leaves or a supplier changes, then handle credentials and data through the company process. A listed application is discoverable, not necessarily authorized.