MCP Servers
Connect any MCP-compatible server once — your voice agents get its tools at the moment a call begins and call them mid-conversation.
Paid plansYour own serverMCP — the Model Context Protocol — is an open standard that lets AI assistants discover and call external tools. If you run an MCP-compatible server (your own, or one of the many community servers), you can plug it into Hanc.AI and your agents will get the server's tools at the moment a call begins.
Once connected, the agent can call those tools mid-conversation — pulling live data, running calculations, kicking off backend workflows — exactly like the built-in Calendar, HubSpot, or API Tool integrations.
MCP servers are available on every paid plan, from Starter — not in the free trial and not on Free; there the MCP card says "Connecting MCP servers is available on paid plans". There's no per-server fee — you only pay for the call minutes your agent uses when invoking the tools.
What your agent can do
- Call your tools mid-conversation — pull live data, run calculations, kick off backend workflows, exactly like the built-in integrations.
- Get the current tool set on every call — tools are discovered when the call starts, so changes on your server show up on the next call.
- Use several servers without clashes — each tool name is prefixed with its connection's label.
When to Use MCP
| Scenario | Why MCP Fits |
|---|---|
| You already run an internal tool server | One connection, all your tools available to every agent |
| You want to share tools across agents | Connect once, attach to as many agents as you like |
| You use community MCP servers | Plug in popular open-source MCP servers without writing custom code |
| Your tool surface changes frequently | The agent re-discovers tools at each call — add/remove tools server-side without touching the dashboard |
If you have a single one-off endpoint, the API Tool is simpler. If you have a set of related tools, or you already maintain an MCP server, this is the right integration.
Connecting an MCP Server
- Open the Integrations page
In the dashboard sidebar, go to Integrations and open the Developers tab: the MCP servers section is at the top.

- Click Connect MCP server
The MCP servers panel lists your connections. Click Connect MCP server.

A modal opens. Fill in:
Field Description Label A short name for this connection (e.g. Sales Tools,Internal CRM). It is shown in the connection list and in the agent's list of MCP servers, and it prefixes the tool names presented to the AI, so keep it short and meaningful.Server URL The HTTPS address of your MCP server's Streamable HTTP endpoint, including the path (often /mcp).http://and private or internal addresses are rejected.Authentication None if your server is open, or Bearer token if it expects an Authorization: Bearer …header.Bearer token (Bearer only) Your token, sent as Authorization: Bearer …. Stored encrypted; Hanc.AI decrypts it only to connect to your server on your behalf. When you edit the connection, leave the field empty to keep the current token.
Click Connect. Hanc.AI accepts only
https://addresses and rejects private and internal hosts. It does not check whether your MCP server answers at that address: a wrong path or a server that is down only shows when the agent can't see the tools.The connection is now available to every agent in your workspace.
Tool names from each MCP server are prefixed with the connection label, so an MCP server with a lookup_order tool connected as "Shop API" will appear to the agent as Shop_API_lookup_order. This prevents collisions when you connect multiple MCP servers that happen to have similarly-named tools.
Edit or delete
The MCP servers list shows the label, URL, authentication and status (active / inactive) of every connection you've added. Use the buttons in each row:
- Edit — change the label, URL, authentication or token, then click Save
- Delete — remove the connection entirely
Removing a connection deactivates it for every agent that had it attached — those agents simply lose access to its tools on the next call.
Attaching MCP Tools to an Agent
A connection in your workspace is available to your agents — but each agent decides individually which of your connections to expose. This means you can have separate Sales and Support agents that pull from different MCP servers.
- Open the agent's Actions tab
Pick the agent, open the Actions tab and click Add Tool.
- Choose MCP servers
The menu groups entries by phase: Pre fetch, Live call, Post call. MCP servers is in the Live call column, under CRM & services — the agent calls MCP tools during the conversation, not before or after it.
- Fill in the tool
Field Description Tool name A name for this entry in the agent's Actions tab (e.g. External MCP tools). Required.When to use it A short instruction the agent reads to decide when to reach for these tools during a conversation. Required. MCP servers Every MCP connection in your workspace, each with an on/off switch. Turn on the ones this agent should use — all tools those servers expose become callable. A connection that is not active is marked inactive. 
Click Save. The agent now has access to the tools from every enabled MCP connection.
You can have one MCP connection enabled on five agents and another enabled on only one — each agent has its own MCP servers tool with its own set of toggles.
Example workflows
Look up an order on your own server
The caller asks where their order is → the agent calls the lookup_order tool of your MCP server, connected as "Shop API" (Shop_API_lookup_order) → it answers from what your server returns.
Different tools for different agents Your Sales agent has the Sales Tools connection switched on, your Support agent the Internal CRM one → each agent sees only the tools of its own connections.
A new tool without touching the dashboard You add a tool on your MCP server → the very next call discovers it → the agent can use it right away.
How It Works at Call Time
When a call starts, the agent connects to each of its enabled MCP servers and asks "what tools do you offer?" Whatever the server returns is exposed to the AI as a tool it can call during the conversation.
A few things to know:
- Tools are discovered at call start. If you add or remove tools on your MCP server, the change shows up on the very next call — no need to update anything in Hanc.AI.
- Tool names are prefixed with the connection label (so
"Shop API"+ toolsearch_products→Shop_API_search_products). Built-in agent tools and other MCP tools won't clash. - Tool descriptions and parameter schemas come from your MCP server. Write good descriptions there — that's what the AI uses to decide when and how to call the tool.
- Errors are reported back to the agent. If a tool call fails, the agent sees the error message and can decide to retry, apologise, or try a different tool.
Requirements for Your MCP Server
To work with Hanc.AI, your server needs to:
| Requirement | Detail |
|---|---|
| HTTPS | Plain http:// is not accepted. Use a real TLS certificate. |
| Public URL | The URL must resolve to a public address — Hanc.AI can't reach servers on a private network or localhost. |
| MCP Streamable HTTP transport | Your server must speak the Streamable HTTP transport on a single endpoint (usually /mcp). Servers that only offer the older SSE transport (an /sse endpoint) and stdio servers are not supported. |
| Bearer auth (optional) | If your server requires authentication, it must accept a standard Authorization: Bearer <token> header. |
If you're hosting your own server behind a firewall, expose it via a tunnel or reverse proxy with a public HTTPS endpoint.
Security and Privacy
Your bearer tokens are stored encrypted at rest. Hanc.AI decrypts a token only to connect to your server on your behalf.
Tool calls flow agent → Hanc.AI → your MCP server → back. Hanc.AI logs tool arguments and the first 2,000 characters of each result for troubleshooting. Your server logs whatever you choose to log on your end.
If you change or rotate your token, edit the connection in the dashboard — the new value takes effect on the next call.
To let your server know which signed-in user of your website is calling, have your server issue a short-lived token for that user and pass it through the website widget. It arrives at every MCP connection of the agent in the X-Hanc-Support-Token header — see Identifying a Signed-In User.
Plan Availability
| Plan | MCP Servers |
|---|---|
| Free trial, Free | Not available |
| Starter / Pro / Business / Enterprise | Available |
There is no per-connection charge — you only pay for the call minutes during which the agent invokes the tools.
Troubleshooting
"Connection won't save"
The dashboard shows Could not save the connection. Please check the values and try again. Check that:
- The URL begins with
https://, nothttp://. - Check the address for typos; private and internal hosts are rejected.
- The host is public —
localhost,127.0.0.1and internal addresses are blocked by design.
"Agent doesn't see the tools"
- Check the address: Hanc.AI doesn't check that your MCP server answers there. Make sure the URL points to your server's Streamable HTTP endpoint, path included (often
/mcp), and that the server is running. - Check that the agent's MCP servers tool has at least one connection switched on, that the connection is active, and that the tool is saved.
- Make sure your MCP server answers a tools/list request without an interactive sign-in.
- Re-check the bearer token — extra spaces or quotes make your server reject it.
"Tool fails mid-call"
- The agent will see whatever error your server returns. Check your server logs.
- Long-running tools may time out — keep MCP tool execution under a few seconds per call so the visitor doesn't sit in silence.
"I changed tools on my server but the agent still uses old ones"
- Tools are re-discovered at the start of each call, not cached across calls. The very next new call sees the updated tool set. Ongoing calls keep the tools they were given at start.
Related
- Actions & Tools — Built-in tools and how the Actions tab is organized
- Integrations — All available integrations
- API Tool RAG — Simpler option for a single endpoint