MCP Servers
MCP (Model Context Protocol) servers let you extend your assistant with custom tools and integrations.
What is MCP?
MCP is a protocol for connecting AI models to external tools and data sources. With MCP servers, you can:
- Add custom tools to your workflows
- Connect to internal APIs
- Integrate with services not built into Town
- Extend capabilities without code changes
How it works
- You configure an MCP server in Settings
- When creating/editing a workflow, you can enable that MCP server
- The workflow gains access to the MCP server’s tools
- During runs, your assistant can call those tools
Configuring MCP servers
Add a server
- Go to Settings → MCP
- Click Add Server
- Enter server details:
- Name — Display name
- URL — Server endpoint
- Description — What this server provides
Server requirements
MCP servers must:
- Implement the MCP protocol
- Be accessible from Town servers
- Respond with tool definitions
Enable for workflows
When creating or editing a workflow:
- Scroll to the MCP Servers section
- Check the servers you want to enable
- The workflow now has access to those tools
Using MCP tools
Workflows use MCP tools like any other tool:
- Available tools are shown in the workflow’s tool list
- Your assistant decides when to call them
- Results are returned to the workflow
You can see MCP tool calls in the run logs.
Example use cases
Internal APIs
Connect to your company’s internal APIs:
- Query your CRM
- Update ticket systems
- Access internal databases
Custom integrations
Integrate services not built into Town:
- Specialized SaaS tools
- Industry-specific platforms
- Proprietary systems
Enhanced capabilities
Add specialized capabilities:
- Custom data processing
- Specialized search
- Domain-specific analysis
MCP server info in workflows
When you enable an MCP server for a workflow, you’ll see:
- Server name
- Summary of capabilities
- Tool count — Number of available tools
- Estimated tokens — Context size impact
MCP servers add to context size. Each server’s tools are included in the system prompt, consuming tokens.
Best practices
Only enable what you need. Each MCP server adds to context size. Only enable servers a workflow will actually use.
Test MCP connections. Before enabling for production workflows, test that the MCP server is accessible and working.
Monitor for failures. Check run logs to ensure MCP tool calls are succeeding.
Resources
- MCP Protocol Documentation
- Building your own MCP server (coming soon)