# Tools and MCP

The tools agents can call, the policies that decide who may call them, and how Katonic connects to MCP clients and servers.

A **tool** is an action an agent can take beyond answering from a model: search a knowledge source, call an HTTP endpoint, run code, or use a capability exposed by an MCP server. This section explains the tool catalog, tool policies, and Katonic's two roles in the Model Context Protocol: as a host that runs MCP servers from its server catalog or your own, and as a server that exposes platform tools to MCP clients.

**Who can use it:** builders attach tools to agents; administrators decide, through tool policies, which tools may be called and by whom.

Video: Tools and MCP (3:04), https://katonicdocsmedia.blob.core.windows.net/docs/tools-and-mcp/dc3542776092/index.html

## The Tools page, tab by tab

The video above walks the Control Room's **Tools** page, where every source of tools is registered, inspected and governed. Each tab in a step; the pages in this section go deeper.

  
**Step 1. Servers**

Open **Tools** under **Agents & AI** in the Control Room sidebar. The health chip reads **All healthy** or names what needs attention; chips count servers and databases, and **Import OpenAPI** and **Add MCP Server** add a source. **Servers** lists every MCP server and OpenAPI source, one card each; the chips filter by kind (**Deployed**, an image on your cluster; **Hosted**, a server elsewhere; **OpenAPI**, an imported REST spec) and **Search servers** narrows by name. A card shows the auth, the tool count, calls over 24 hours and a status pill, **Ready** when agents can use it; the switch starts and stops a server, or enables an OpenAPI source, and **Fix Auth** replaces it when a credential has expired. Select a card for its detail: status, type, auth, transport, groups, and every discovered tool with its risk. **Databases** lists connections: the engine, its tools, **Read-only** when only curated queries run, and a status.

  
**Step 2. All tools**

**All tools** answers what your agents can actually do: one row per tool with **Source**, **Kind**, **Risk** and **Basis**; chips filter by source kind, the segment by risk. **Basis** says how the risk was decided: **Declared by server**, **From HTTP method**, or **Inferred from name**. **Inferred only** isolates the guesses, and a tool whose source cannot serve a call is marked **Unavailable**.

  
**Step 3. Server catalog**

**Server catalog** is the MCP catalog synced from Ops Hub. **Deploy** opens a card's configuration, a deployed card shows its status, and **Refresh Catalog** pulls the current entries.

  
**Step 4. Built-in**

**Built-in** lists the tools that run inside Katonic, each with an availability switch. Where one cannot work, say **Web Search** without outbound network, switch it off with a reason: it reads **Disabled** and a reason field appears. Select **Save changes**; a notice confirms, and in the agent wizard the tool stays listed, locked and marked **Unavailable**.

  
**Step 5. Add a source**

**Import OpenAPI** takes a spec from a URL, pasted JSON or a file, then **Configure**; [Connect an API as a tool](/tools/connect-an-api-as-a-tool) has the walkthrough. **Add MCP Server** takes a Docker container for the cluster or a remote, hosted server, with its transport, authentication, resources and access groups; [Add an MCP server](/tools/add-an-mcp-server) covers it.

## In this section

  - [The tool catalog](https://docs.katonic.ai/tools/tool-catalog): The Tools page in the Control Room: servers, databases, the inventory, the Ops Hub catalog and the built-in tools.
  - [Add an MCP server](https://docs.katonic.ai/tools/add-an-mcp-server): Add a hosted or self-hosted MCP server from the server catalog or from Tools, choose how it authenticates, and manage it.
  - [Connect an API as a tool](https://docs.katonic.ai/tools/connect-an-api-as-a-tool): Import an OpenAPI spec so agents can call the API, with identity forwarding, rate limits and caching.
  - [Host your own tool server](https://docs.katonic.ai/tools/host-your-own-tool-server): Two ways to put code you wrote in front of agents on your own cluster, and how to choose between them.
  - [Connect an MCP client to the Katonic MCP Server](https://docs.katonic.ai/tools/katonic-mcp-server): Point Claude Desktop, an MCP SDK client, OpenAI Agents or OpenCode at the platform's built-in tools with one API key.

Tool policies and the approvals they trigger are covered in [Govern the platform day to day](/get-started/govern-day-to-day) and [Automations and approvals](/user-guide/work-room/automations-and-approvals).

## Related

- [Agents](/agents/overview)
- [Automations and approvals](/user-guide/work-room/automations-and-approvals)

Source: https://docs.katonic.ai/tools/overview
