Skip to content
AnthropicAnthropic

Building with MCP and the Claude API

Anthropic’s Alex Albert (Claude Relations), John Welsh (Engineering, MCP team), and Michael Cohen (Engineering, Claude API team) discuss the origins of the Model Context Protocol (MCP), the open standard for connecting AI applications to external systems, best practices for getting started with MCP, and how MCP and Claude work together to enable more powerful agentic systems. 00:00 - Introductions 00:30 - What is MCP? 1:30 - The origins of MCP 2:50 - Open sourcing MCP 5:00 - Remote MCP support 6:15: MCP registries 7:40 - Favorite MCPs: Context7 & Playwright 10:40 - Using the Claude API MCP connector 11:50 - Prompt engineering with MCP 14:20 - Best practices for managing context and tools with MCP 18:20 - How John and Michael use MCP servers for project management, home automation, and more 20:00 - Understanding the “emergent” behaviors when Claude and MCP servers work together 22:50 - The future of MCP: growth of the protocol and ecosystem Learn more about MCP: https://modelcontextprotocol.io/docs/getting-started/intro Learn more about the Claude Developer Platform: https://www.claude.com/platform/api Learn more about how to write effective tools for AI agents: https://www.anthropic.com/engineering/writing-tools-for-agents

Alex AlberthostMichael CohenguestJohn Welshguest
Oct 8, 202525mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

MCP connects Claude to tools, context, and real-world actions.

  1. MCP is a universal protocol for giving Claude external context and tool access beyond the chat transcript, enabling real-world actions like web access or app integrations.
  2. Anthropic built MCP to avoid repeatedly re-implementing the same tool integrations across products, aiming for “build once, configure everywhere.”
  3. Open-sourcing MCP was intended to create an industry-wide standard so vendors don’t need separate connectors per model provider, accelerating ecosystem growth.
  4. Remote MCP support and emerging registries reduce setup friction and increase trust by enabling official, hosted endpoints (e.g., GitHub) instead of locally running random connectors.
  5. Effective MCP use depends on prompt-quality tool design and disciplined context/tool selection to reduce token cost, ambiguity, and model confusion as tool counts grow.

IDEAS WORTH REMEMBERING

5 ideas

Treat MCP tools as part of the prompt, not just an interface.

Tool names, parameter names, and descriptions directly shape model behavior; adding brief guidance and examples can noticeably improve how Claude invokes and uses a tool (e.g., better image-generation prompts).

Remote MCP support is a major unlock for adoption.

Early MCP often required users to run servers locally, which was clunky and blocked vendors from hosting official integrations; remote-hosted servers make onboarding much faster for end users.

Registries reduce the “random connector” trust problem.

Instead of installing an unknown community connector locally, developers can increasingly point Claude to an official vendor endpoint (e.g., GitHub) discovered via registries, improving safety and reliability.

Use fewer, better-scoped tools to avoid confusion and cost.

Stuffing many servers/tools into a request increases tokens and can confuse the model—especially when multiple tools overlap (e.g., Linear and Asana both exposing “Get Project Status”).

Design MCP servers differently than traditional APIs.

LLMs can handle “fuzzier” intent-driven interfaces; often 1–2 higher-level tools can outperform 15–20 granular endpoints because the model can infer needed parameters from descriptions.

WORDS WORTH SAVING

5 quotes

MCP is, uh, the Model Context Protocol, and, uh, it's a way of providing external context to models.

John Welsh

So build it once and configure everywhere.

Michael Cohen

One of the things that we had noticed is that models having access to external context is, is kind of good for everyone. It's like a rising tide floats all boats type of situation.

John Welsh

MCP servers and tools are really, at its core, prompts.

John Welsh

I, yeah, I'm never, I'm never actually writing anything anymore. It's just all Claude.

Michael Cohen

Definition of MCP as external context/tool protocolMotivation: reuse across Claude surfacesOpen standards and ecosystem incentivesRemote MCP servers and easier setupMCP server registries and “official” endpointsNotable MCPs: Context7 and PlaywrightClaude API native MCP connectorTool/prompt design: names, descriptions, examplesContext window management and tool scopingEmergent behaviors from combining MCP serversFuture: competition on “best MCP server” quality

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome