#model-context-protocol
62 episodes
#5520: Keeping a Fine-Tune Alive Across Model Releases
Daniel wants to fine-tune DeepSeek Flash 4.1 on edited podcast scripts. The hard part isn't training — it's surviving the next release.
#5467: Small Models, Big Schemas: When JSON Constraints Backfire
Small models plus strict JSON schemas should be a safe bet. A 15,000-generation study found the opposite.
#5463: What Happens When You Click "Connect" on an AI Plugin
You click accept, the tool runs, and a credential you never saw is doing the work. Here's who actually owns it.
#5455: Custom GPTs Are Sunsetting: What Happens to Your Actions
OpenAI is retiring Custom GPTs — and custom Actions don't survive the migration. What that means for your workflow.
#5208: Why Tech Consortia Keep the Stack From Flying Apart
From GSM to HDMI to LTO, why competitors form standards bodies — and what happens when the commitments break.
#4713: MCP vs Browser-Use: The Agentic AI Showdown
Two patterns are fighting to define how AI agents interact with software. Which one wins shapes the next decade of building.
#4712: Codex vs Claude Code: The Real Agentic Coding Showdown
Benchmarks, model variants, and MCP setups — which coding agent actually wins where it matters?
#4703: MCP Binary Payloads: Why URLs Beat Base64
MCP can't carry raw bytes. Here's why the URL pattern is the only legal move—and what the File Uploads Working Group might change.
#4701: Why Claude Code Re-Introduces Itself to Your Repo
Claude Code remembers your preferences but not your codebase. We explore why that's by design and how to work around it.
#4700: Browser Agents: Three Paths to One Protocol
How screenshot watching, dev-tools extensions, and native MCP in Chrome are converging — and what wins.
#4663: Python vs TypeScript for MCP Servers: The Real Decision Rule
Both are first-class for MCP, but the real choice depends on what your server talks to. Here's the framework.
#4582: Designing a Home MCP Gateway for Conversational AI
How to securely connect ChatGPT to a home-hosted MCP gateway — transport, auth, and the missing pieces.
#4580: The Tiny AI Tools That Beat Big Agents
Why small, single-purpose AI utilities outperform autonomous agents for daily work — and what interface they need.
#4522: The Sixth Shim: When MCP Becomes a Middleware Factory
Six servers, six shims, no architecture. A taxonomy of MCP integration pain.
#4521: MCP Transport Architecture in 2026: Gateway Design
Three transports, one gateway. How to bridge stdio, legacy SSE, and streamable HTTP in production.
#4167: Why AI Can’t Read Your Gmail (Yet)
Your AI misses emails and fails to send replies. It’s not dumb—the pipes are leaky.
#3755: Hermes vs OpenClaw: Mobile-to-Server AI Frameworks
Why developers are leaving OpenClaw for Hermes—and why mobile-to-server AI interaction remains unsolved.
#3271: LLMs as Parsers, Not Calculators
Stop letting LLMs do math. Use them to parse messy text, then let deterministic code handle the numbers.
#2690: Where Agent Builders Actually Gather
The MCP community, A2A protocol, and Linux Foundation are building the professional identity of agentic AI right now.
#2689: Why CLI Beats MCP for AI Agents Sometimes
Why a plain command-line tool can outperform a purpose-built MCP server for AI agents — and what that means for the protocol's future.
#2683: MCP vs Agent Skills: Context Wars
When 12M token windows arrive, do MCP servers or agent skills win? Plus: federated access for agent teams.
#2551: How Progressive Disclosure Saves MCP from Token Bloat
Why dumping all tool schemas into context breaks accuracy — and three implementations that fix it.
#2541: Agent-to-Agent Scheduling: Building the Calendly for AI
How Google's A2A protocol and Anthropic's Remote MCP could power a new kind of agent handoff for scheduling meetings.
#2478: MCP File Handling: Why Your Base64 Upload Breaks at 4MB
MCP has no standard file input. Base64 breaks at 4MB, presigned URLs need whitelisting, and MinIO workarounds aren't standardized.