Implement Advanced Tool Use with Google ADK based MCP Client

Complete implementation of Programmatic Tool Execution (MCP) with Google ADK & Gemini along with full code. I explain the difference between Direct (per-call) tool invocation and Programmatic Tool Calling (the agent writes/executes code to orchestrate multiple MCP tools inside a sandbox), and demonstrate both using a working example built with Google’s Agent Development Kit (ADK). You’ll see why programmatic execution reduces context/token usage, lowers latency for multi-tool workflows, and makes complex multi-step tasks deterministic and auditable. ----------------------------- Discounted Google ADK Udemy Course (get completion certificate, Q&A, bite-sized lectures) https://theailanguage.com/best_price?course=adk (with instructor discount) ----------------------------- CODE IS AVAILABLE NOW - *Please subscribe, allow pop-ups, and log in to access our GitHub repos.* 👉 Code used in this video (version_7): https://theailanguage.com/onlySubscribers?id=adk_samples_vid_ewGEbeWzWI0_list_PL6tW9BrhiPTAZts0W5nQS9dbW6VMnLKab&site=github MCP STREAMABLE HTTP SERVER CODE Streamable HTTP MCP Servers Used from Repository https://theailanguage.com/onlySubscribers?id=mcp_streamable_http_vid_ewGEbeWzWI0_list_PL6tW9BrhiPTAZts0W5nQS9dbW6VMnLKab&site=github Refer to main/streamable_http_server/1-stateless-streamable/server1.py AND main/streamable_http_server/1-stateless-streamable/server2.py ⸻ If you found this explanation or demo useful: 1. LIKE the video — that helps me decide whether to release the full cleaned code early. 2. COMMENT below saying “I want the code” and what part you’d like documented first (agent setup, MCP server, sandbox runner, or cleanup fix). I track demand and will prioritize accordingly. 3. SUBSCRIBE and enable notifications if you’re building agents — I post code walkthroughs and fixes regularly. Key features covered • Tool Search Tool, Programmatic Tool Calling, Tool Use Examples - Anthropic’s “Advanced Tool Use” announcement (Nov 24, 2025) and what it means for multi-tool agents.  • How programmatic tool calling lets an agent generate a Python script that calls many MCP tools inside a sandbox — reducing the number of model round-trips.  • Practical ADK implementation details and demo (Agent Development Kit).  • Model Context Protocol (MCP) basics and tool metadata/schema. Why this matters (short) Programmatic execution shifts tool orchestration out of the model’s context window and into a sandboxed runtime. That means: • far fewer tokens used for multi-step tool workflows, • less latency from repeated model round-trips, and • clearer auditing & security boundaries for tool access. Who is this for Developers and engineers building agentic systems, integration engineers working with MCP or ADK, and researchers exploring multi-tool agent orchestration. ⸻ My website / code & updates: https://theailanguage.com ⸻ CHAPTERS 00:00 Introduction 01:37 What is Direct Tool Calling 04:55 What is Programmatic Tool Calling 10:27 Demo: Programmatic Tool Calling with Google ADK 15:48 Demo: Direct Tool Calling (comparison) 18:03 Code - see description for link 18:33 Outro ⸻ Code & downloads • Code will be published on my website (theailanguage.com) in 2–3 days. I’m finishing a resource-cleanup fix - I’ll post the cleaned-up repo link in a pinned comment and update this description when it’s live. • If you want the code as soon as it’s ready, leave a like and drop a comment below. I read every comment and will prioritize sharing the implementation if there’s demand. ⸻ #programmatictoolcalling,#advancedtooluse,#mcp,#ai,#agents,#modelcontextprotocol,#anthropic,#claude,#googleadk,#agentdevelopmentkit,#aidev