Tag

#mcp

25 posts

Security & PrivacyOctober 22, 202615 min
Claude Code MCP Permission Rules: 2.1.299 Tested
Claude Code 2.1.299 changed which MCP servers an allow or deny rule covers. I ran 2.1.296, 2.1.299 and 2.1.307 with servers named a and a__b. Here is what ran.
AI & Machine LearningOctober 6, 202617 min
MCP Token Audience Check: expectedResource Tested (2026)
I tested the MCP TypeScript SDK's new expectedResource option for token audience validation. It is off by default and Express adapter 2.0.1 ignores it (2026).
LLM & IntegrationSeptember 28, 202619 min
Claude Inline Tools Beta: What Actually Ships (2026)
Claude's inline tools beta adds tools mid-conversation without busting the prompt cache. I traced exactly when each SDK shipped it, wire format included.
Security & PrivacySeptember 23, 202616 min
AI SDK Tool Drift Detection: What It Misses in 2026
I tested AI SDK v7 fingerprintTools and detectToolDrift against a real MCP server. It catches description and schema rug pulls, but not annotation drift.
AI & Machine LearningSeptember 18, 202619 min
MCP Tasks Extension: A Working TypeScript Server (2026)
The MCP Tasks extension defines three methods for async tool calls, but the official SDK still ships an older shape. Here is a tested server that gets it right.
NewsSeptember 3, 202615 min
Model Hardware Standard: Anthropic's MHS Preview 2026
Anthropic's Model Hardware Standard lets AI agents drive real lab and factory instruments. What MHS is, what the six partners measured, and what it is not.
NewsSeptember 1, 202617 min
Agentic Resource Discovery: AWS Backs the ARD Spec in 2026
AWS now backs Agentic Resource Discovery, the open ARD spec for how AI agents find MCP servers and skills — how ard.json works, and what v0.91 changed.
NewsAugust 31, 202623 min
Web Search for AI Agents: Prices and Indexes in 2026
Bing's search API is gone and Google's ends January 1, 2027. What agent-native web search costs in 2026, from $1 to $8 per thousand queries, and what it hides.
AI & Machine LearningAugust 26, 202615 min
AI Agent Reliability Benchmark 2026: 65% Once, 25% Always
ThinkingBox is a new AI agent reliability benchmark: the best model scores 65.36% pass@1 but only 25.25% pass^20 across 507 stateful business agent tasks.
NewsAugust 25, 202613 min
A2A Joins AAIF: The 2026 Agent Protocol Consolidation
A2A is joining the Agentic AI Foundation, announced August 17, 2026, placing agent-to-agent messaging beside MCP. What the consolidation means for agent teams.
AI & Machine LearningAugust 20, 202637 min
MCP Server Not Showing Up in Claude Desktop: Fixes (2026)
MCP server not showing up in Claude Desktop? Check the Connectors menu, the config's JSON, a full quit, mcp.log, and the file the app reads — in that order.
Tooling & ProductivityAugust 13, 202614 min
Agent Plugins 1.0: One Plugin, Nine Clients in 2026
Agent Plugins 1.0 packages Agent Skills and MCP servers into one portable folder that loads in nine AI clients. Vercel holds the spec's deciding vote.

Every post in #mcp (25)