Connect external AI tools, services, and data sources to your Composer workflows through MCP (Model Context Protocol). MCP Servers Overview

​
What is MCP?

MCP lets you extend Composer with:
  • External AI services like Claude, ChatGPT, or custom models
  • Third-party tools that aren’t built into Composer
  • Internal services and APIs from your organization
  • Data sources like databases, file systems, or knowledge bases

​
Quick setup

  1. Go to Settings → Integrations → MCP Servers
  2. Add a new server with the connection details
  3. Test the connection to make sure it works
  4. Grant permissions for what the server can access
  5. Use in workflows - agents can now call these tools
Add MCP Server Form Configure MCP servers by specifying the server name, transport type (Server-Sent Events recommended), and server URL endpoint.

​
Common MCP integrations

AI models:
  • Connect custom LLMs or specialized AI services
  • Use domain-specific models (legal, medical, financial)
  • Integrate proprietary AI tools your company has built
Data sources:
  • Connect internal databases and knowledge bases
  • Access file systems, document repositories
  • Integrate with CRM, ERP, or other business systems
Development tools:
  • Custom CI/CD systems not in the built-in catalog
  • Internal code analysis or security scanning tools
  • Proprietary deployment or monitoring systems

​
Using MCP in workflows

Once connected, MCP tools work like any other tool: In chat: “Use our custom security scanner to check the main branch” In workflows: Add MCP tools as workflow steps alongside built-in tools. For agents: Agents automatically know about MCP tools and can use them.

​
Security and permissions

Network isolation: MCP servers run in isolated environments. Permission scoping: Control what each MCP server can access. Audit logging: All MCP tool usage is logged and trackable. Credential management: API keys and secrets are stored securely.

​
Troubleshooting

Connection issues: Check network access and firewall settings. Authentication failures: Verify API keys and credentials are current. Performance problems: Monitor resource usage and optimize queries. Tool not appearing: Check server registration and permission settings.

​
Advanced configuration

Custom protocols: Extend MCP with your own communication protocols. Load balancing: Distribute requests across multiple server instances. Caching: Improve performance with response caching. Monitoring: Set up alerts for server health and performance.
New to MCP? Start with a read-only integration like a database or file system to get familiar with how it works.