HomeProduct › MCP
Tool Layer

A full Model Context Protocol client.

MCP is the emerging standard for exposing tools to AI systems. AGNT speaks it natively — point it at any MCP server and those tools immediately become callable by every agent and every workflow you have built.

Download free MCP setup docs
Why It Matters

The connector list stops being a ceiling

Cloud automation platforms are bounded by their catalog: if the vendor has not built an integration, you wait. MCP inverts that — anyone can publish a server, and any MCP-speaking client can use it the day it ships.

01 / OPEN

Not a walled garden

MCP servers are written by tool vendors, open-source maintainers, and internal teams. None of them need AGNT's permission, and AGNT does not need theirs.

02 / PRIVATE

Run your own

Expose an internal system to your agents by writing a small MCP server behind your firewall. Nothing about it has to become public to be usable.

03 / PORTABLE

No lock-in

Tools built against a standard move with you. Work invested in an MCP server is not stranded if you change your agent platform later.

Four Ways to Extend

MCP is one of several routes

Reach for the lightest one that solves the problem. Most teams end up using all four.

Native nodes

119+ built-in tools

Filesystem, HTTP, database, code execution, and every native integration. Zero setup — they are present the moment you install.

MCP servers

The open ecosystem

Connect any MCP server, local or remote. Its tools are then discoverable and callable by agents exactly like native ones.

Custom tools

Prompt-defined tools

Define a tool from a prompt and an input schema, no code required. Useful for repeatable reasoning steps you want named and reusable.

Plugins

Full code extensions

Ship real code with its own dependencies, UI, and nodes. Signed, permission-scoped, and publishable to the marketplace.

Safety

Third-party tools, first-party control

Explicit permissions

Connecting a server does not hand it to every agent. Tool access is granted deliberately, per agent.

  • Tools assigned per agent, not globally
  • Plugins declare permissions before install
  • Permission changes on update require re-approval
  • Trust badges on marketplace listings

Gated execution

Every tool call passes a security gate before it runs — including calls originating from third-party servers.

  • Prompt-injection detection on tool inputs
  • Secret-leak scanning on tool outputs
  • Dangerous-command interception
  • Human approval on high-consequence actions