Skip to main content

AI MCP

Bridges an MCP server to the Ai.ToolProvider contract, so an Ai.Agent can discover and call a whole MCP server's tools.

Why use this​

  • Discovery, not declaration — point an agent at a server and it gets every advertised tool; no per-tool manifest entry.
  • Agent stays MCP-agnostic — AiMcp.ToolProvider implements the generic Ai.ToolProvider abstract; @telorun/ai never depends on MCP.
  • Transport-pluggable — wraps any Mcp.Client (Mcp.StdioClient, Mcp.HttpClient).
  • The only coupling point — this is the single module that depends on both @telorun/ai and @telorun/mcp-client; the two cores stay single-purpose.

Kinds​

KindPurpose
AiMcp.ToolProviderAn Ai.ToolProvider backed by an Mcp.Client (tools/list + tools/call).

Example​

kind: Telo.Application
metadata: { name: my-app, version: 1.0.0 }
secrets:
  openaiApiKey:
    env: OPENAI_API_KEY
    type: string
imports:
  Ai: std/ai@^0.4.0
  AiOpenai: std/ai-openai@^0.4.0
  Mcp: std/mcp-client@^0.4.0
  AiMcp: std/ai-mcp@^0.4.0
---
kind: AiOpenai.OpenaiModel
metadata: { name: Gpt4o }
model: gpt-4o-mini
apiKey: "${{ secrets.openaiApiKey }}"
---
kind: Mcp.StdioClient
metadata: { name: FilesMcp }
command: npx
args: ["-y", "@modelcontextprotocol/server-filesystem", "/data"]
clientInfo: { name: my-app, version: 1.0.0 }
---
kind: AiMcp.ToolProvider
metadata: { name: FileTools }
client: !ref FilesMcp
---
kind: Ai.Agent
metadata: { name: Assistant }
model: !ref Gpt4o
toolProviders:
  - provider: !ref FileTools
    prefix: "fs_"

Reference​