Start here
No connector requiredTry the job without a connector
The agent can use local files, websites, and desktop controls. A connector is useful when the job needs a service-specific tool.
A Skill can tell Think-Act how your team triages a support ticket. A connector can give it a ticket lookup if your service offers one. With an exported ticket list, it can try the routine without connecting anything.
Windows 10 and 11 now. Mac release planned.
Integration workflow
Try the built-in tools first. Add a connector when the missing piece is access to a specific service.
The smaller workflow is usually easier to understand
External capability is useful when it is necessary. It should not become a default dependency simply because a connector exists.
Start here
No connector requiredThe agent can use local files, websites, and desktop controls. A connector is useful when the job needs a service-specific tool.
Add when needed
External capabilityIf a ticket service exposes a lookup through MCP, enable that lookup and try it on one ticket you know.
The task genuinely depends on another supported system or tool.
The capability is actually configured and available to the workflow.
Its source, access, scope, and consequences are understood well enough to use deliberately.
A Skill describes how to work. It doesn’t grant access to files, accounts, or services.
Skill
A Skill can keep reusable instructions, useful conventions, prerequisites, review points, and expected output structure for a familiar task.
A Skill does not create credentials, permissions, connectors, application access, files, or external authority by itself.
Connector or tool
A connector or tool can expose specific supported capability from another system. Its setup, requirements, access, and behavior depend on the connector involved.
Having a connector configured does not mean every task should use it. Match external access to the workflow and its consequence.
How the pieces relate
Need
Capability gap
Configure
Connector setup
Review
Source & access
Test
Verify behavior
Choose a connector from a source you trust, check its requirements, and test it with one low-risk lookup.
Name what the workflow cannot do yet and why an external tool or system needs to participate.
Setup may involve a runtime, credentials, application access, network access, local resources, or other connector-specific requirements. Do not assume one universal configuration.
Begin with a bounded task and confirm the returned result or external effect before making the workflow broader or more consequential.
Third-party MCP packages need independent review
Third-party packages are not controlled or audited by PixLab unless explicitly stated. Review the source, maintainer, configuration, commands, credentials, requested access, and real behavior before trusting one with meaningful work.
For ticket triage, save which queue to check, what counts as a blocker, and when to ask a person. Supply fresh tickets on the next run.
Preserve how the task should be approached, including prerequisites and decision points that should remain understandable next time.
Keep the review points, warnings, and expected result visible so reuse does not remove the safeguards that made the method useful.
Save guidance after you understand the workflow's inputs, expected output, required capabilities, boundaries, and useful checks.
A Skill does not create
Method
Reusable guidance
Checks
Review points
Reuse
After testing
A local MCP package can execute code on your machine. Review its source, permissions, and data access before enabling it.
State should be understandable without color
Use clear language such as Configured, Trusted, Available, or Permitted rather than relying on a green indicator alone.
A useful mental model, not a promise of one universal internal state machine.
Configured
The relevant connector or tool needs to exist in the workflow's supported configuration.
Trusted
Review third-party source, requirements, access, and credentials before relying on the connector.
Available
Do not plan the result around a capability that is missing or unavailable when the task runs.
Permitted
Match access and confirmation to the action and consequence instead of treating every tool call the same.
A connector may receive meaningful access depending on its design and configuration. Review it before trusting it with real work.
If a required capability cannot be used, surface the limitation rather than treating the external action as successfully completed.
Ask for a specific result. If a service is unavailable, the agent should report the missing access rather than invent its answer.
Desktop Agent
Runs the job using the context and supported capabilities available to the workflow.
Agent detailsIntegrations
Adds configured external tools only when the job cannot be completed with the capabilities already available.
Automate Tasks
Turns an understood workflow into something repeatable where the method and safeguards are stable enough.
Automation detailsA Desktop Agent task can remain connector-free. A connected task does not have to become an automation. Combine these pieces only when the work benefits from the combination.
Try one read-only lookup before using a connector in a longer workflow.
Keep the first workflow narrow
One task and one clearly needed capability are easier to understand, test, and review than a broad integration setup.
Begin
Start with writing, OCR and documents, screen recording, or a bounded task based on supplied context.
Identify
Name the system the task must reach and what Think-Act actually needs to do there.
Review access
Review its requirements, credentials, permissions, and the systems it may interact with before trusting it with real work.
Test
Check the connector's behavior, returned result, and relevant external effect before increasing the consequence of the task.
Control
Use permissions and confirmation appropriate to the action, then review important external changes.
Maintain
Revisit integrations as work changes and remove access that no longer belongs in the workflow.
Use the connectors shown in your installation. Don’t assume an enterprise integration or remote agent service is included.
Planned / roadmap
Broader managed integration concepts may include cloud connector discovery, organization-level policies or provisioning, and remote-agent integration ideas.
What they do, when you need them, and how their permissions work.
No. Start with built-in tools. Add a connector when you can name the service or action that is missing.
A Skill is a recipe for a task. A connector gives the agent an additional tool; saving a recipe does not grant access.
No. Review the publisher, package source, and permissions of a third-party connector before running it.
That step cannot run through the missing connector. Check the configuration or ask the agent to prepare a draft instead.
No. A Skill can describe the steps, but the external tool must exist and be permitted.
You can use local MCP connectors and reusable Skills with the agent's built-in tools.
Don't plan a rollout around future features. The integration described here is local MCP, not a hosted connector registry or a remote agent fleet.
You can try Think-Act on 64-bit Windows 10 or 11. The Mac release is planned for Q1 2027; see Downloads for availability. See downloads page and compare plans.
Test the instructions on one exported ticket. Add a trusted lookup only if reading the live queue would save a step next time.