---
title: "Think-Act Integrations | Local MCP, Skills And Enterprise Connectors"
description: "Connect Think-Act to trusted tools with local MCP servers, skills, and planned enterprise Cloud MCP and A2A remote-agent connectors."
---

# Think-Act Integrations

Think-Act integrations are designed around a simple rule: the agent can use only the tools that are installed, trusted, enabled, and healthy.

Skills shape behavior. Connectors provide real capabilities.

## Integration Model

Think-Act separates guidance from action.

- Skills tell the agent how to approach a workflow.
- MCP tools let the agent call real capabilities.
- Security policy decides whether actions are allowed, blocked, or need approval.
- Failed connectors are disabled instead of exposed.
- The agent should recommend missing connectors instead of inventing tools.

## Local MCP Connector

Local MCP servers run on the user device.

- They may require Node.js, npx, Python, or another runtime.
- They can access files, credentials, networks, or apps exposed to them.
- They should be installed only from trusted sources.
- They can be enabled, disabled, tested through use, and removed.
- Their tools are passed to the agent only when the server is enabled and connected.

## Marketplace And Custom Servers

Users can add tools from two paths.

- Marketplace search for compatible stdio MCP packages.
- Manual configuration for custom local servers.
- Duplicate packages are blocked while installed.
- Required fields are kept minimal.
- Trust confirmation is shown before installation.

## Skills

Skills are local playbooks.

- Built-in skills support research, review, planning, writing, security, and productivity.
- Imported skills should be reviewed before use.
- Skills can require connectors for real actions.
- Safety modes can keep skills manual, read-only, or connector-required.
- Skills should not promise actions that no tool can perform.

## Enterprise Connectors

Enterprise connector work is planned after local MCP is stable.

- Cloud MCP connector registry.
- Paid connector gating.
- Managed auth and team policies.
- Enable and disable controls.
- Tool exposure to the agent through the same safety model.
- A2A remote agents after the connector layer is proven.

## Security Guidance

Treat integrations as privileged access.

- Use trusted packages only.
- Keep write-capable tools behind confirmation.
- Avoid passing secrets as plain environment variables where possible.
- Disable connectors when not needed.
- Review logs after failed or suspicious tool calls.

## Get Started

Start with writing and OCR, then add one trusted connector.

[Download Think-Act now](./downloads.md) and connect tools only when a workflow clearly needs them.

## Related Pages

- [Home](./index.md)
- [Security](./security.md)
- [Desktop Agent](./feature-desktop-agent.md)
- [Automate tasks](./automate-tasks.md)
- [Features](./features.md)
- [Downloads](./downloads.md)

## FAQ

### Are local MCP packages controlled by PixLab?

No. Third-party local MCP packages are not controlled or audited by PixLab. Install only what you trust.

### Are Cloud MCP connectors available in this release?

No. They are planned enterprise features and should be marked as not implemented for this release.

### Can skills call tools directly?

Skills guide the agent. Real tool calls require an enabled connector that exposes the needed tool.