👾 3 Ways to Integrate MCP Servers in Your Agentic Workflows

Welcome to Edition #9 of Agents Made Simple

This week was a bit quieter regarding new releases. A chance to catch up with existing technology and try new things out.

For instance, you could create a smart assistant using Mistral's new developer kit, integrate MCP tools into your current processes, or use a Gemma-powered agent on your computer or phone to automate privacy tasks.

This week’s topics:

• Mistral’s new Agents API

• Google DeepMind’s new Gemma models

• Japan’s self-improving agent

• 3 ways to integrate MCP servers in your agentic workflows

• Plus AI investments, trending AI tools, community highlights, and more

AI Agent News Roundup

💥 Breakthroughs

📈 Investments

🇺🇸 xAI led by Elon Musk agreed on a deal with Telegram, bringing the Grok chatbot to over a billion Telegram users in exchange for $300M and a revenue-share.

🇺🇸 The New York Times announced an AI licensing deal with Amazon, granting the tech giant permission to use its editorial content to train its AI models.

🇨🇳 Nvidia is about to produce a cheaper Blackwell AI chip specifically for the Chinese market in response to recent U.S. export curbs.

🇦🇪 The United Arab Emirates secured the ChatGPT Plus ($20/m) subscription for its entire population, becoming the first nation to offer premium AI services for no cost.

10x Your Outbound With Our AI BDR

Scaling fast but need more support? Our AI BDR Ava enables you to grow your team without increasing headcount.

Ava operates within the Artisan platform, which consolidates every tool you need for outbound:

• 300M+ High-Quality B2B Prospects, including E-Commerce and Local Business Leads

• Automated Lead Enrichment With 10+ Data Sources

• Full Email Deliverability Management

• Multi-Channel Outreach Across Email & LinkedIn

• Human-Level Personalization

Book a demo and supercharge your sales team.

3 Ways to Integrate MCP Servers in Your Agentic Workflows

Difference Between MCP and Function Calling Tools

Before the release of the Model Context Protocol, you had to implement tools (or functions) yourself. What’s the innovation? In simple terms:

Function Calling = AI decides what and when to call a tool, calling it directly.

MCP = A universal system that standardizes how tools are discovered, accessed, and managed, acting like a marketplace of AI tools.

This means MCP offers more scalability, flexibility, and easier integration for multiple tools, while normal function calls offer simplicity and speed for straightforward tasks.

No-Code Examples

1. Make.com: You can set up MCP servers and clients in Make. To set up an MCP server, you first need to create an MCP token in your user’s settings. Then paste it in the MCP configuration for your MCP client.

   Your client (e.g., Cursor, Claude Desktop, or programmed agents) can then call any Make scenario you configured to run on demand.
   Example config for Claude Desktop:

{
  "mcpServers": {
    "make": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://<MAKE_ZONE>/mcp/api/v1/u/<MCP_TOKEN>/sse"
      ]
    }
  }
}

2. n8n: Supports both MCP server and client nodes. The MCP server node allows n8n to function as an MCP server by exposing workflows as tools accessible to MCP clients.

   The MCP client node enables n8n to connect to external MCP servers, allowing it to use the capabilities of those servers within its workflows.

Advanced Production-Ready Integrations

3. CrewAI: Supports the integration of MCP tools. Install the extra package:

uv pip install 'crewai-tools[mcp]'

Here is an example Python script that integrates the ScrapeGraphAI MCP server in CrewAI for agents to call:

import os
from crewai import Agent
from crewai_tools import MCPServerAdapter
from mcp import StdioServerParameters

# Replace this with your actual ScrapeGraphAI API key
SCRAPEGRAPH_API_KEY = "YOUR-SGAI-API-KEY"

# Define the MCP Stdio server parameters
server_params = StdioServerParameters(
    command="npx",
    args=[
        "-y",
        "@smithery/cli@latest",
        "run",
        "@ScrapeGraphAI/scrapegraph-mcp",
        "--config",
        f'{{"scrapegraphApiKey":"{SCRAPEGRAPH_API_KEY}"}}'
    ],
    env={**os.environ}
)

# Use the MCP tools in a CrewAI agent
with MCPServerAdapter(server_params) as mcp_tools:
    print(f"✅ Loaded tools: {[tool.name for tool in mcp_tools]}")

    agent = Agent(
        role="Web Data Harvester",
        goal="Scrape and analyze web data using ScrapeGraphAI tools.",
        backstory="Expert at leveraging ScrapeGraphAI for data extraction.",
        tools=mcp_tools,
        reasoning=True,
        verbose=True
    )

    # Example usage of the agent (replace with your Crew setup)
    # For example, you could assign this agent a task in a Crew workflow.

In other agent frameworks, the MCP integration works similarly. It’s basically a plug-and-play way to make external software available to your agents.

Tool Spotlight

✨ Run AI offline on your phone

You can run small LLMs like Gemma3 or Qwen2.5 by installing the Google AI Edge Gallery app. Currently only available on Android.

How it works:

1. Download the .apk file from the official GitHub repo releases page to your device.

2. Open the APK file on your device to install it.

3. Now you can download the available models to your device and try them out. Depending on your hardware, not all models will run fluently. I recommend starting with the smallest Gemma3-1B-IT q4 variant.

🎥 KlingAI: Kling 2.1 launched to compete directly with Google's Veo 3. Generate cinematics with significant improvements in generation speeds, prompt adherence, realism, and reduced artifacts.

📷 FLUX.1 Kontext: A text-to-image model that performs in-context image generation, allowing you to prompt with both text and images, and seamlessly extract and modify visual concepts to produce new, coherent renderings.

🗣️ Rime: A text-to-speech language model that offers AI voices with personality and emotions. Ready for your agents via API with <200ms in latency.

🗣️ Hume EVI3: A speech-language model that can understand and generate any human voice and personality from a prompt in <1s. It uses a voice-to-voice architecture and comes with a deeper understanding of tune, rhythm, timbre, and speaking style.

Community Highlights

1 Trick to Stop and Correct Hallucinations in Agentic Systems

How it works:

• Let the agent tell you when it needs more information

• In the response format, add a parameter where the agent can give a complaint to you, the developer (e.g., call it debug_info)

• The agent should use that parameter when it detects confusing or unspecified information, and it doesn’t know what to do

• Run the LLM in production with real user data and go back and look at the outputs

• It ends up being a to-do list that the agent developer has to tackle to improve the agentic system over time

More Resources

Blog: In-depth articles on AI workflows and practical strategies for growth
AI Tool Collection: Discover and compare validated AI solutions
Consultancy: I help you discover AI potential or train your team