🔌 MCP
Model Context Protocol — how clients (Claude, Cursor, IDEs) connect to servers that expose tools, resources, and prompts.
The workflow
flowchart LR
A[MCP Client<br/>Claude, Cursor, IDE] --> B[Discover server<br/>capabilities]
B --> C{Primitive}
C -->|Tools| D[Callable functions<br/>read / write / act]
C -->|Resources| E[Static / dynamic<br/>context]
C -->|Prompts| F[Reusable<br/>templates]
D --> G[Invoke + return<br/>structured result]
E --> G
F --> G
MCP standardizes how AI clients connect to anything. Build once, plug into every client.
Key takeaways
Videos (42)
Building Agents with Model Context Protocol - Full Workshop with Mahesh Murag of Anthropic
MCP turns custom AI integrations into a standardized client-server protocol so any compatible client can talk to any tool with zero glue code.
The Future of MCP — David Soria Parra, Anthropic
MCP is becoming the universal context protocol — moving beyond tools to ship full apps that both humans and models interact with.
MCP is all you need — Samuel Colvin, Pydantic
Use MCP — especially sampling — as a one-protocol-fits-all way for autonomous Python agents to compose tools and share LLM access.
A2A & MCP Workshop: Automating Business Processes with LLMs — Damien Murphy, Bench
A2A handles remote agent-to-agent discovery while MCP handles tool/context plumbing — pick them only when you don't control both sides.
MCP = Mega Context Problem - Matt Carey
Treat MCP tool catalogs as a retrieval problem, not a dump — load only the relevant tools per turn to keep context usable.
Gateways are All You Need — Karan Sampath, Anthropic
For enterprise MCP adoption to scale, security teams must bless a single gateway that handles auth, observability and deployment so every team can ship MCP servers safely.
Your MCP Server is Bad (and you should feel bad) - Jeremiah Lowin, Prefect
Treat an MCP server as a curated product UI for agents, not a 1:1 mapping of your existing API surface.
MCP UI: Extending the frontier — Liad Yosef and Ido Salomon, MCP Apps
MCP Apps standardizes returning interactive, branded UI from MCP tools, decomposing the web into agent-composable atoms that keep the host model in control.
MCP: Origins and Requests For Startups — Theodora Chu, Model Context Protocol PM, Anthropic
MCP went from internal Anthropic hack to multi-lab standard by optimizing for model agency, server simplicity, and bidirectional agent communication.
The agent-ready web: Simplify user actions with WebMCP — Tara Agyemang, Google
Expose clear website actions as structured tools while maintaining strong accessibility and ordinary browser usability.
MCPs are Boring (or: Why we are losing the Sparkle of LLMs) - Manuel Odendahl
MCP is great plumbing but standardizing every LLM interaction as tool-calling discards the language-generative magic that made LLMs interesting.
Combine Skills and MCP to Close the Context Gap — Pedro Rodrigues, Supabase
MCP supplies access, while carefully structured skills supply the product knowledge needed to use that access correctly.
Building Interactive UIs in VS Code with MCP Apps — Marlene Mhangami & Liam Hampton, GitHub
MCP Apps combine tool execution with direct manipulation interfaces that keep users inside their current workflow.
Building Agent Interfaces: Lessons from Chrome DevTools (MCP) for Agents — Michael Hablich, Google
Design tools around agent reasoning limits and measure successful outcomes rather than simply minimizing tool count or tokens.
Letting AI Interface with your App with MCP — Kent C Dodds
MCP is the standard glue layer that finally lets every app expose itself to every AI assistant — build MCP servers or get bypassed.
Remote MCPs: What we learned from shipping — John Welsh, Anthropic
Inside a large org, every custom tool API converges on MCP — adopt it as the default context interface even before you need all features.
MCP Is Not Good Yet — David Cramer, Sentry
MCP is a useful plug-in architecture for agents but you must design tool surfaces for LLMs, not just wrap your existing API — and OAuth on remote MCP is the only sane path for B2B SaaS.
Why MCP and ChatGPT Apps Use Double Iframes — Frédéric Barthelet, Alpic
Embedding interactive MCP views requires balancing host isolation, content-security policy, and app storage capabilities.
Exposing Agents as MCP servers with mcp-agent: Sarmad Qadri
Treat agents as MCP servers and async workflows — you get composable multi-agent systems for free over standard protocols, with no monolithic framework.
Your Agent's Biggest Lie: "I Searched the Web" — Rafael Levi, Bright Data
Check actual retrieval evidence and tool access before trusting an agent's claim that an answer reflects the live web.
From MCP to Scale: Pipelines That Build Themselves — Rafael Levi, Bright Data
Use models for pipeline construction and repair, and deterministic scripts for repeated extraction at scale.
The rise of the agentic economy on the shoulders of MCP — Jan Curn, Apify
MCP's tool discovery enables agent-to-tool interaction at scale, but a real agentic economy requires solving agent identity and payments — likely via marketplaces before central rails mature.
Blender MCP and The Future Of Creative Tools - Siddharth Ahuja
Scriptable creative tools plus MCP plus AI asset generators collapse historically complex 3D workflows from hours into minutes of prompting.
MCP Tasks (async): Why Aren't Any Agents Supporting Them? — Cornelia Davis, Temporal
Long-running MCP tools require durable coordination on the client as well as the server.
MCP Apps: Extending the Frontier — Ido Salomon & Liad Yosef
MCP Apps preserve rich application interactions while keeping the AI host in control of tool execution.
MCP Apps: Primitives, discovery, and the Future of Software - Pietro Zullo, Manufact, Inc
Useful MCP Apps combine interactive UI with explicit state synchronization and accessible distribution.
Real world MCPs in GitHub Copilot Agent Mode — Jon Peck, Microsoft
Copilot Agent Mode + MCPs makes IDE agents capable of complex multi-tool tasks while keeping the developer in the permission loop.
Building the Engine While Flying the Plane: Launching the Figma MCP Server — Jesse Lumarie, Figma
Design-to-code integrations need structured visual context, real component reuse, and repeatable evaluation.
Introducing WebMCP: Agents in the Browser — RL Nabors
Existing web infrastructure can expose content and interactions to agents alongside ordinary browser users.
Are MCPs Overhyped? A Rant about MCPs — Henry Mao, Smithery
MCP solves LLM-service connectivity but reputation, install friction, monetization, and discovery remain — Smithery aims to be the AI gateway for that ecosystem.
Full Spec MCP: Hidden Capabilities of the MCP spec — Harald Kirschner, Microsoft/VSCode
Full MCP spec — sampling, resources, dynamic discovery, roots — unlocks rich stateful agent UX far beyond the tool-only patterns most servers ship today.
One Login to Rule Them All: Cross-App Access for MCP — Garrett Galow, WorkOS
Cross-App Access via the corporate IDP eliminates per-MCP consent screens and restores enterprise revocation/visibility that today's MCP OAuth flow breaks.
One Registry to Rule them All - Sonny Merla, Mauro Luchetti, & Mattia Redaelli, Quantyca
An AI gateway plus MCP/A2A/use-case registries gives a large enterprise the governance, cost-attribution and lineage it needs to actually scale agentic AI safely.
Scaling GitHub for your Agents — Sam Morrow, GitHub
MCP servers at platform scale need aggressive tool reduction, server-side multi-call workflows, eval-driven tool descriptions, and OAuth instead of plaintext PATs.
MCP Agent Fine tuning Workshop - Ronan McGovern
Bootstrap a stronger MCP agent by running Qwen-3 + Playwright MCP, saving multi-turn reasoning traces, and fine-tuning on its own successful runs.
(possible dupe but better sound) What does Enterprise Ready MCP mean? — Tobin South, WorkOS
Enterprise-ready MCP requires bolting standard identity, OAuth scopes and audit layers onto remote MCP servers, including multi-hop agent delegation.
The State of MCP observability: Observable.tools — Alex Volkov and Benjamin Eckel, W&B and Dylibso
Treat MCP-powered agents as a distributed system and propagate OpenTelemetry context through MCP's meta payload to recover end-to-end tracing.
Your Insecure MCP Server Won't Survive Production — Tun Shwe, Lenses
Good MCP design and good MCP security are the same discipline — get the interface wrong and OAuth can't save you.
Just do it. (let your tools think for themselves) - Robert Chandler
Make MCP tools agentic — give them natural-language inputs and let them reason internally — instead of exposing thin API wrappers that overflow the parent agent's context.
Bending a Public MCP Server Without Breaking It — Nimrod Hauser, Baz
You can dramatically improve third-party MCP tool behavior by rewriting descriptions, scoping, and wrapping them — without forking the upstream server.
Building Protected MCP Servers — Den Delimarsky and Julia Kasper, MCP Steering Committee & Microsoft
Remote MCP servers should sit behind a proper identity provider with OAuth/PKCE and per-tool scopes — treat them like any other protected enterprise API.
Agents, Access, and the Future of Machine Identity — Nick Nisi (WorkOS) + Lizzie Siegle (Cloudflare)
MCP plus OAuth on top of Cloudflare Durable Objects gives agents real, scoped, persistent machine identity rather than ad hoc API keys.