Anthropic Built a Fix That Proves the Problem 🔍

Right, so Anthropic dropped Tool Search on November 24th alongside Claude Opus 4.5, and I need you to sit with the implications for a second because they're brutal for MCP.

Tool Search lets you mark tools as defer_loading: true. When you do that, the tool's schema never enters your context window until Claude actually needs it. Claude gets a name and nothing else. When a task comes in that requires the tool, Claude calls the Tool Search tool, pulls the schema, and only then does it have the full definition loaded. Lazy loading for AI tools. Sounds boring. It's not.

Here's the number that should make you spit your tea out: accuracy went from 49% to 74% on Opus 4. On Opus 4.5, it climbed from 79.5% to 88.1%. Same model. Same tools. The only difference is not loading the tool definitions upfront.

I'll say it again. The mere presence of tool schemas in context was tanking accuracy by 25 percentage points. Twenty-five. Not because the model got confused by irrelevant tools (though it did). Because the context window was full of JSON nobody was using, and the model was drowning in it.

---

This is an admission dressed up as a feature

I've been banging on about MCP's context cost since July. The Playwright MCP server alone burns 12.8k tokens just sitting there. Load four or five MCP servers and you've torched half your context window before asking the model to do anything useful.

Tool Search is Anthropic's answer. And the answer is: don't load the tools.

Think about that for a second. The company that created MCP built a feature whose entire purpose is avoiding the cost of loading MCP tool definitions. If that isn't an admission that the protocol has a context problem, I don't know what is. They didn't say "we've made MCP more efficient." They said "here's a way to not load MCP schemas until the last possible moment." The fix is the diagnosis.

And the timing is proper interesting. Three weeks earlier, on November 5th, mcporter dropped. mcporter converts MCP servers into plain code. Regular functions. No protocol overhead, no JSON schemas burning context, just code your agent calls directly. Two completely different approaches to the same problem, landing in the same month. The ecosystem is routing around MCP overhead from both directions simultaneously: Anthropic from the top (defer the loading) and the community from the bottom (skip the protocol entirely).

📚 Geek Corner

Deferred loading vs. elimination: Tool Search and mcporter solve the same problem differently. Tool Search keeps MCP but hides it until needed (lazy loading). mcporter removes MCP entirely and converts tool definitions to native code (compilation). The lazy loading approach preserves the MCP ecosystem but still pays the schema cost when tools activate. The compilation approach pays zero ongoing cost but loses MCP's dynamic discovery. Both are concessions that the original "load everything upfront" model was broken.

---

The real question nobody's asking

If deferred loading improves accuracy by 25 percentage points, what does that tell us about every benchmark result published before November 24th? All those MCP-heavy setups were being evaluated with bloated context. Accuracy scores were lower than they needed to be. And some of those "model X is better than model Y" comparisons? Potentially measuring context pollution as much as model capability.

I reckon we're going to see a wave of "actually, turns out our setup was fine, we just had too many tool schemas loaded" revelations over the next few months. The MCP ecosystem built a house of cards where adding more tools made every tool worse, and nobody measured it because the degradation was gradual.

Feels like: Discovering your car's been running with the handbrake on for six months. The engine was fine the whole time. You were just dragging unnecessary weight.

Bottom line: Anthropic built Tool Search because MCP's context cost is a real and measured problem. Fewer tools in context means better accuracy, full stop. If you're running more than a couple of MCP servers, either defer them or convert them. And if you want the full argument for why MCP is heading for the bin, the Death of MCP piece lays it out end to end. The protocol had a good run. The ecosystem is moving on.