MCP Servers and AI Tools
What are MCP servers?
MCP (Model Context Protocol) servers give your Velaro AI bots access to real-time tools — database lookups, API calls, order status checks, and more. Instead of hard-coding every integration into a workflow, you connect an MCP server once and the AI bot can call its tools naturally in conversation.
Example: a visitor asks "What is the status of order #8821?" The bot calls the get_order tool on your MCP server, retrieves the live order data, and replies — all without an agent.
Connecting an MCP server
- Go to MCP Servers in the sidebar (under Integrations).
- Click Add MCP Server.
- Enter:
- Name — a label for your team (e.g. "Order Lookup API")
- URL — the HTTPS endpoint of your MCP server
- API Key — if your server requires authentication
- Click Test Connection — Velaro will discover the available tools.
- Click Save.
The tools on this server are now available to any AI bot in your account.
Using MCP tools in a bot
- Open the AI Bots section.
- Edit the bot that should use the tools.
- Under Tools, toggle on the MCP server(s) you want this bot to use.
- In the system prompt, describe when and how the bot should call the tools. Example:
When a customer asks about an order, use the get_order tool with their order number.
Always confirm the order number before calling the tool.
- Save and activate the bot.
Building your own MCP server
An MCP server is any HTTPS endpoint that responds to the MCP protocol. Velaro-compatible servers expose a tool manifest at GET /tools and accept tool calls at POST /tools/{toolName}.
Minimal Node.js example:
const express = require('express');
const app = express();
app.use(express.json());
app.get('/tools', (req, res) => {
res.json({ tools: [
{
name: 'get_order',
description: 'Look up an order by order number',
parameters: {
type: 'object',
properties: {
orderNumber: { type: 'string', description: 'The order number' }
},
required: ['orderNumber']
}
}
]});
});
app.post('/tools/get_order', async (req, res) => {
const { orderNumber } = req.body;
// fetch from your database
const order = await db.orders.findOne({ id: orderNumber });
res.json({ result: order });
});
app.listen(3000);
Host this on any cloud provider (Vercel, Railway, AWS Lambda, Azure Functions) and add the URL in Velaro.
Security best practices
- Always use HTTPS — Velaro will not connect to plain HTTP endpoints.
- Use an API key to restrict access to your server to Velaro only.
- Return only the data the bot needs — do not expose entire database records.
- Set rate limits on your MCP server to prevent runaway tool calls.
Tool call limits
Your plan includes a limit on MCP tool calls per conversation and per month. These limits prevent runaway loops and control costs. You can see current usage in Settings > Billing. Contact your account manager to increase limits.
AI Orchestration (multi-bot delegation)
With orchestration enabled, one AI bot can delegate to another. For example, your main customer service bot can hand off to a specialized "technical support bot" for complex questions, then resume when it's done.
To set up delegation, your primary bot's system prompt should include:
If the customer has a billing question, delegate to the billing_specialist bot.
If the customer has a technical question, delegate to the technical_support bot.
Velaro handles the handoff automatically. The conversation history is preserved across both bots.
Troubleshooting MCP connections
"Connection failed" — check that the server URL is correct, uses HTTPS, and is publicly accessible. Verify the API key.
"No tools found" — ensure your server's GET /tools endpoint returns a valid tool manifest.
"Tool call timed out" — your MCP server must respond within 15 seconds. Optimize slow database queries or add caching.
Tools not being called — add explicit instructions to the bot's system prompt telling it when and how to use each tool.
Was this article helpful?