The most common question we get from IT and operations buyers isn't "can your AI answer questions?" It's: "We already have an AI agent running in Azure / Copilot Studio / our own API — do we have to rebuild everything to use your chat platform?"
The answer is no. Velaro is designed to connect to external AI systems, not replace them. Whether you have a knowledge base in SharePoint, a custom agent in Azure AI Foundry, or a Copilot Studio bot your team built last quarter, there's a direct integration path. This post covers the three patterns, when to use each, and the specific Microsoft path in detail.
The Three Integration Patterns
Every AI integration into Velaro chat workflows falls into one of three categories. The right one depends on whether your data is static or live, how latency-sensitive the lookup is, and how much structure your AI's response has.
KB Ingestion — Ingest Once, Search Always
You point Velaro's Knowledge Base at your content source — a URL, a sitemap, uploaded PDFs, or a set of articles — and Velaro indexes it into a vector search index. During a conversation, when the AI needs to answer a question, it searches that index semantically and returns the most relevant content to ground its response.
This is the lowest-latency, lowest-cost path for content that doesn't change frequently. The AI doesn't call your systems in real time — it reads from the prebuilt index.
- Your content is product documentation, FAQs, policy pages, or knowledge articles
- The information changes weekly or monthly, not per-customer or per-request
- You want search to be instant with no external API dependency at conversation time
- You're starting a migration and want something running in under a day
MCP Server — Tool Call Mid-Conversation
MCP (Model Context Protocol) lets you expose your AI agent as a callable tool. Velaro's AI can invoke it mid-conversation, pass structured inputs, and receive structured outputs that it then uses to form its response. The call happens in real time — the customer's question triggers it, your agent processes it, and the result comes back into the conversation thread.
This is the pattern for anything that requires live data: current inventory, a customer's account status, open support tickets, order details that weren't available at indexing time.
- You need the response to reflect the current state of a live system
- Your existing agent already has logic you don't want to duplicate in Velaro
- The data is customer-specific and changes frequently (account balances, order status, entitlements)
- You want structured input/output contracts between Velaro and your agent
HTTP Node — Any REST Endpoint, No Wiring Required
The HTTP Node is a workflow step that makes an HTTP request to any REST endpoint and passes the response into the next step. It doesn't care whether the endpoint is an AI agent, a CRM lookup, a pricing calculator, or a custom Lambda function. You configure the URL, headers, request body template, and a JSON path to extract the value you care about. That value becomes a variable in the rest of the workflow.
- Your target system has a REST API but isn't MCP-compatible
- You need a simple lookup with a deterministic response shape
- You want to integrate Power Automate or Logic Apps flows into a chat workflow
- You're prototyping — HTTP Node is the fastest path from idea to working integration
When to Use Each Pattern: A Decision Guide
| Signal | Best Pattern | Why |
|---|---|---|
| Static product / FAQ content | KB Ingestion | Index once, zero per-query cost, instant retrieval |
| Customer-specific data (orders, accounts) | MCP or HTTP Node | Must be live — KB can't reflect per-customer state |
| Existing agent with complex logic | MCP Server | Preserves your agent's logic, adds structured contract |
| Simple REST lookup, known response shape | HTTP Node | Faster to configure, no MCP wrapper needed |
| Power Automate / Logic Apps trigger | HTTP Node | Logic Apps expose HTTP triggers natively |
| Hybrid: static docs + live lookups | KB + MCP (both) | KB for product info, MCP for account data — each workflow step uses the right source |
Want to see how your existing AI stack connects to Velaro? Talk to an integration specialist — no sales pitch, just a technical walkthrough.
Talk to an integration specialistThe Microsoft Path: Copilot Studio, Azure AI Foundry, and Power Automate
Microsoft's AI ecosystem is the most common starting point for enterprise customers. Here's the specific path for each tool.
Copilot Studio: Publish as MCP Server
Microsoft added native MCP server publishing to Copilot Studio in 2025. If your team has already built a Copilot Studio agent, you can publish it as an MCP endpoint without rebuilding anything. The setup from a customer's side takes roughly 30 minutes:
-
Publish your Copilot Studio agent as an MCP server In Copilot Studio, navigate to Publish → Channels → MCP. Enable the MCP endpoint and copy the generated URL and bearer token. Microsoft handles the protocol translation — your agent's existing topics and actions become MCP tools automatically.
-
Configure the MCP connection in Velaro In Velaro Admin, go to Integrations → MCP Servers and add a new server entry. Paste the Copilot Studio endpoint URL and bearer token. Velaro will discover the available tools from the MCP manifest and list them for use in workflow nodes.
-
Wire the tool call into your workflow In the Workflow Editor, add an AI Node and select your Copilot Studio tool from the tool picker. Set the input mapping (what conversation context to pass) and the output variable name. The agent's response flows into subsequent workflow steps.
-
Structure your agent's output for downstream use Copilot Studio agents return a text response by default. For downstream workflow logic — conditional branches, variable extraction, data mapping — you want structured JSON. Add a response formatting step to your Copilot Studio agent's topic that outputs a JSON object rather than a plain text string. Velaro's HTTP Node can then extract specific fields using a JSON path expression.
What "MCP export" actually produces from Copilot Studio
Copilot Studio's MCP export wraps your agent's topics as tool definitions with input schemas. The agent still executes its full topic logic, including any Power Platform connectors, Dataverse lookups, or Azure OpenAI calls you've configured inside Copilot Studio. Velaro calls the MCP endpoint, your Copilot Studio agent runs, and returns a result — from Velaro's perspective, it's a structured tool call.
The important caveat: the response is unstructured text by default. If you want Velaro's AI to extract specific fields and take conditional action based on the response, you need to modify your Copilot Studio agent to return a JSON object with named fields. This is a one-time change to the agent's response format topic, not a change to Velaro.
Azure AI Foundry: HTTP or MCP
Azure AI Foundry agents expose HTTP endpoints via the Foundry API. You have two options:
HTTP Node (simpler setup): Call the Foundry agent endpoint directly from a Velaro HTTP Node. You'll need the endpoint URL, your Azure API key or managed identity token, and the request shape that the Foundry agent expects. The HTTP Node sends the customer's message or extracted context, the agent responds, and Velaro extracts the response field via a JSON path.
MCP (structured contract): If you deploy the Foundry agent behind an MCP-compatible wrapper (Azure provides an MCP adapter in preview), you get tool discovery, input validation, and structured output without writing a custom JSON path for each field. This is worth the extra setup if you're exposing multiple tools from the same agent or if the response shape changes frequently.
Power Automate and Logic Apps
Power Automate flows and Logic Apps both support HTTP triggers — an endpoint that your workflow fires when an HTTP POST arrives. This makes them straightforward to wire into Velaro's HTTP Node:
- Create a Power Automate flow with an HTTP trigger as the entry point
- Configure the flow to run your existing actions (SharePoint lookups, Teams messages, Dynamics queries, Approvals)
- Add a Response action at the end that returns a JSON body with the data Velaro needs
- Paste the trigger URL into Velaro's HTTP Node, add the authorization header, and map the response fields
The HTTP trigger URL includes an embedded SAS token that acts as authentication. You don't need to configure OAuth or manage credentials separately — just copy the URL from Power Automate and paste it into Velaro.
A Realistic Integration Scenario
Here's how these patterns work together for a mid-size enterprise customer running Microsoft's stack:
A customer asks in chat: "What's my current SLA entitlement and do I have any open tickets against it?"
The workflow runs like this: Velaro's AI first checks the KB index (which contains the customer's service tier documentation, indexed from a SharePoint library) to understand what the SLA entitlements for their tier look like. It then calls a Copilot Studio agent via MCP, passing the customer's account ID. The Copilot Studio agent queries Dynamics 365 and returns the customer's current SLA usage and any open tickets. Velaro combines both results — the general SLA definition from KB and the customer's specific usage from the live lookup — and forms a response that answers both parts of the question in a single reply.
Neither pattern alone would have answered the full question. The KB has the policy; the live tool call has the customer-specific data. This is the combination that gets deflection rates above 70% for complex, account-specific queries — not just FAQ coverage.
What You Don't Need to Do
It's worth being explicit about what this integration model does not require:
- You don't need to rebuild your AI agent inside Velaro's platform. Your existing Copilot Studio agent, Foundry model, or internal API stays exactly as-is. Velaro calls it.
- You don't need to duplicate your knowledge base if it's already in SharePoint or Confluence. Velaro's KB ingestion pipeline can index it directly from the source URL.
- You don't need to change your authentication infrastructure. Velaro's HTTP Node supports Bearer tokens, API keys, and OAuth 2.0 client credentials. Your existing auth scheme works.
- You don't need Velaro to be the system of record. Velaro is the conversation layer and orchestration engine. Your data stays in your systems. Velaro reads it when a conversation needs it.
Claude + Velaro: One Command Away
If your team uses Claude — Claude Code for engineering, Claude Desktop for operations, or Claude via the API for internal tooling — Velaro ships a native MCP server that connects Claude directly to your support infrastructure. No wrapper code. No custom authentication. One command:
npx @velaro/mcp-server
# With your key
VELARO_MCP_KEY=vel_live_... npx @velaro/mcp-server
Add it to Claude Code's ~/.claude/settings.json and Claude has direct operator access to your knowledge base, can search and create articles, and can query your account data — all from inside your AI coding sessions or day-to-day Claude Desktop conversations.
npx invocation always pulls the latest version. No package.json maintenance required. The hosted HTTP endpoint at api-admin-us-east.velaro.com/mcp also works with Claude Desktop 0.8+ for clients that prefer not to run a local Node process.This is the gap that separates Velaro from the rest of the market on this specific dimension: Intercom, Zendesk, and Freshdesk do not publish MCP servers. Connecting Claude to those platforms requires building a custom MCP wrapper against their REST APIs, handling authentication yourself, and maintaining the integration as their APIs change. Velaro ships the server, hosts it, and keeps it updated. The integration is one npx command.
Getting Started
The fastest path to a working integration is to start with the HTTP Node. Pick one lookup your team does manually in chat today — account status, order details, ticket lookup — find the REST endpoint that answers it, and wire it into a Velaro workflow. A working prototype is typically a half-day of configuration, not a development project.
Once the HTTP Node is working and you understand the response shape, you can decide whether to graduate it to a full MCP integration for better tool discovery and input validation, or leave it as-is if the HTTP Node covers everything you need.
KB ingestion is a parallel workstream, not a prerequisite. Index your product documentation while the HTTP Node integration is being configured. By the time the live lookup is working, you'll have a KB index ready to combine with it.
For teams already using Claude, the @velaro/mcp-server npm package is the fastest onramp. One MCP key from Settings → API → MCP Keys, one line in your Claude config, and Claude can manage your entire knowledge base directly from your chat interface.
Frequently Asked Questions
Can I connect my existing AI agent to Velaro without rebuilding it?
Yes. Velaro supports three integration patterns that work with any AI system that has an HTTP endpoint: KB Ingestion (index your agent's output or documentation once), MCP Server (call your agent as a live tool mid-conversation), and HTTP Node (call any REST endpoint directly from a workflow step). No Velaro code changes are required.
How do I connect Microsoft Copilot Studio to Velaro?
Copilot Studio agents can be published as MCP servers — Microsoft added this capability in 2025. Once published, you point Velaro's MCP integration at the endpoint and configure the authentication token. Setup typically takes 30 minutes for a customer who already has a Copilot Studio agent built.
What is the difference between KB ingestion and an MCP tool call?
KB ingestion indexes content once and makes it searchable at conversation time using vector similarity — it's cheap, fast, and works at scale, but only as fresh as your last reindex. An MCP tool call hits a live endpoint on every customer request, returning real-time data. Use KB for static content, MCP for anything that must reflect current system state.
Does Velaro support Azure AI Foundry agents?
Yes. Azure AI Foundry agents expose HTTP endpoints. You can connect them to Velaro via the HTTP Node (call the endpoint as a workflow step) or via MCP if you deploy the agent behind an MCP-compatible wrapper. The HTTP Node approach requires no MCP configuration — just the endpoint URL, headers, and a JSON path for extracting the response field.
Do I need to change Velaro's code to add a new AI integration?
No. The HTTP Node, MCP integration, and KB ingestion pipeline are all configuration-driven through the Velaro admin UI. You can swap out the backend AI system, point to a different model, or add a new data source without opening a support ticket or waiting for a product release.