Integrations

Framework guides

Integration examples for popular AI frameworks. The pattern is the same across all of them: construct a MiddlewareContext, call proxy.run(ctx, handler), return the result. For stdio servers, the CLI bridge requires zero code changes.

Vercel AI SDKPopular

Use @businys/ops with the Vercel AI SDK tool system.

Vercel AI SDKts

import { createMCPProxy } from "@businys/ops"
import { tool } from "ai"

const proxy = createMCPProxy()

// Wrap your AI SDK tools
async function handleToolCall(toolName: string, args: unknown) {
  const ctx = {
    toolName,
    toolGroup: toolName.split("_")[0] ?? toolName,
    toolTier: "kit" as const,
    method: "tools/call",
    path: "/tools/call",
    input: args as Record<string, unknown>,
    agentId: "vercel-ai-agent",
    serverName: "my-server",
    startedAt: Date.now(),
    destructive: false,
  }

  return proxy.run(ctx, async () => {
    const result = await yourActualTool(args)
    return { content: [{ type: "text", text: JSON.stringify(result) }] }
  })
}

Mastra

Add full ops middleware to any Mastra MCP server.

Mastrats

import { createMCPProxy, observe } from "@businys/ops"

// Wrap Mastra's MCP handler
const proxy = createMCPProxy()
const ops = await observe({ port: 3100 })

// In your Mastra MCP server's tool handler
export async function handleTool(toolName: string, input: unknown, agentId: string) {
  const ctx = {
    toolName,
    toolGroup: toolName.split("_")[0] ?? toolName,
    toolTier: "kit" as const,
    method: "tools/call",
    path: "/tools/call",
    input: input as Record<string, unknown>,
    agentId,
    serverName: "mastra-server",
    startedAt: Date.now(),
    destructive: false,
  }
  return proxy.run(ctx, () => mastraToolHandler(toolName, input))
}

OpenAI Agents SDK

Bridge any OpenAI Agents SDK tool server with full middleware.

OpenAI Agents SDKsh/yaml

# Use the CLI bridge — no code changes to your server
npx @businys/ops bridge node ./openai-agents-server.js --port 3100

# Point your agent at http://localhost:3100 instead of the server directly

LangChain

Wrap LangChain tools with rate limiting, audit logging, and observability.

LangChaints

import { createMCPProxy } from "@businys/ops"

const proxy = createMCPProxy({
  rateLimit: { globalMax: 200, windowMs: 60_000 },
})

// Wrap any LangChain tool.call
const originalCall = tool.call.bind(tool)
tool.call = async (input, config) => {
  const ctx = {
    toolName: tool.name,
    toolGroup: tool.name.split("_")[0] ?? tool.name,
    toolTier: "kit" as const,
    method: "tools/call",
    path: "/tools/call",
    input: input as Record<string, unknown>,
    agentId: config?.runId ?? "langchain-agent",
    serverName: "langchain",
    startedAt: Date.now(),
    destructive: false,
  }
  const result = await proxy.run(ctx, () => originalCall(input, config))
  return JSON.parse(result.content[0]?.text ?? "{}")
}

Any stdio serverNo code required

Wrap any stdio MCP server — Claude, GPT-4, or custom — as a managed HTTP endpoint.

Any stdio serversh/yaml

# Zero-code bridge for any stdio MCP server
npx @businys/ops bridge node ./my-stdio-server.js

# With options
npx @businys/ops bridge python server.py \
  --port 3200 \
  --rate-limit 50 \
  --name "my-python-server"

# The bridge endpoint at http://localhost:3200 now has:
# - Rate limiting (50 calls/min/agent)
# - Reputation scoring
# - Audit logging
# - Confirmation for destructive calls

GitHub Actions

Run Observer Mode in CI to capture every tool call during tests.

GitHub Actionssh/yaml

# .github/workflows/test.yml
jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4

      - name: Install
        run: npm ci

      - name: Start Observer Mode
        run: npx @businys/ops observe --port 3100 &

      - name: Run tests
        run: npm test
        env:
          MCP_OPS_URL: http://localhost:3100

      - name: Print call stats
        run: npx @businys/ops status --port 3100