AppTable
MCP-native · works with Claude Desktop, Code, Routines

Your personal app suite.
Built with Claude.

From chat to structured storage, available as a live app.

See it in motion ↓

Notion is static. Lovable is locked.
AppTable lets your agent build the exact apps you need.

Each app has a database, a frontend, and skills Claude can update. The examples below are real conversations and the working apps they produced.

“Keep me up to date on AI.”

A daily Routine scans your sources, ranks what matters, and files a briefing before you wake up. No tabs, no feeds, no manual curation.

claude.ai
Claude · Opus 4.7
via MCP
apptable.cc/u/sarah/ai-feedprivate

AI Feed

7 picks · filed 07:15
Wed · May 6 · 07:15
Releases3
A

Claude Opus 4.7 released — 1M context, native MCP

First model with MCP tool use baked into the trainer; agent benchmarks jump 11pts.

anthropic.com·2h
O

OpenAI ships Agents SDK 3 with parallel tool calls

Mostly catches up to Anthropic's MCP — but the parallel tool dispatcher is genuinely faster.

openai.com·6h
D

DeepMind: AlphaProof at IMO 2026 takes silver

Solved 4 of 6 problems unaided — informal-to-formal translation step is the new bottleneck.

deepmind.com·1d
Research2
P

Compositional reasoning via tool-use chains

Chains of 5+ tool calls degrade fast — they propose a re-planning checkpoint that keeps accuracy stable to 12.

arXiv 2604.01182·8h
P

On the limits of MCP server scaling

Empirical: tool-routing latency dominates past ~40 connected servers. Caching helps; tool-name hashing helps more.

Anthropic Research·1d
Threads2
K

the fastest way to debug an agent is to literally print every tool call it makes — most …

Worth reading — overlaps with what you wrote about MCP debugging two weeks ago.

@karpathy·5h
S

thread: 6 patterns I've seen for shipping LLM eval suites that don't rot in a month

Pattern 4 (golden-set drift) is exactly what you flagged at the offsite.

@swyx·11h