MCP Integration

Model Context Protocol — connect your agents to external tools and data sources.

AgentOven integrates with the Model Context Protocol (MCP) — an open standard for connecting LLMs to external tools and data sources. MCP lets your agents call APIs, query databases, and interact with services through a unified protocol.

How MCP Works in AgentOven

AgentOven includes a built-in MCP Tool Server that exposes registered tools to agents. Agents can discover and call tools without framework-specific integrations.

Agent receives a task via A2A
Agent queries the MCP Tool Server for available tools
MCP returns tool schemas (name, description, parameters)
Agent selects and calls tools via MCP protocol
Tool results flow back to the agent for reasoning

Registering Tools

Tools are defined as MCP-compatible schemas and registered with the control plane:

yaml

tools:
  - name: web-search
    description: "Search the web for information"
    parameters:
      type: object
      properties:
        query:
          type: string
          description: "Search query"
        max_results:
          type: integer
          default: 5
      required: ["query"]

  - name: database-query
    description: "Query the analytics database"
    parameters:
      type: object
      properties:
        sql:
          type: string
          description: "SQL query to execute"
      required: ["sql"]

Using Tools in Python

python

from agentoven import AgentOven

oven = AgentOven("http://localhost:8080")

# List available MCP tools
tools = oven.mcp.list_tools()
for tool in tools:
    print(f"{tool.name}: {tool.description}")

# Call a tool directly
result = oven.mcp.call_tool(
    name="web-search",
    arguments={"query": "AI safety research 2025", "max_results": 3},
)
for item in result.content:
    print(item.text)

Using Tools in TypeScript

typescript

import { AgentOven } from "@agentoven/sdk";

const oven = new AgentOven("http://localhost:8080");

// List tools
const tools = await oven.mcp.listTools();
tools.forEach((t) => console.log(t.name, "—", t.description));

// Call a tool
const result = await oven.mcp.callTool({
  name: "web-search",
  arguments: { query: "AI safety research 2025", maxResults: 3 },
});
console.log(result.content);

Built-in Tool Servers

AgentOven ships with several built-in MCP tool servers that can be enabled in your kitchen config:

Tool Server	Description
`web-search`	Search the web via configurable search provider (Bing, Google, Brave)
`file-system`	Read, write, and list files within sandboxed kitchen directories
`http-fetch`	Make HTTP requests to external APIs with configurable auth
`database`	Query PostgreSQL, MySQL, or SQLite databases with read-only access
`code-exec`	Execute Python or JavaScript code in sandboxed containers

Custom Tool Servers

You can connect external MCP-compatible tool servers to AgentOven:

toml

[mcp]
# Built-in tools
enable = ["web-search", "file-system"]

# External MCP servers
[[mcp.servers]]
name = "my-custom-tools"
url = "http://localhost:9090/mcp"
transport = "sse"

[[mcp.servers]]
name = "company-apis"
url = "http://internal-tools:8081/mcp"
transport = "stdio"
command = "npx"
args = ["-y", "@company/mcp-tools"]