Changelog Sep 11, 2026 - Neon

Claimable Neon is for developers building with agents. When your agent is mid-build, the signup, email verification, or API key it needs to provision a database brings everything to a halt. Claimable Neon solves this problem. It lets your agent provision a temporary project with scoped credentials, without an account or payment details, and keep building. If you like the result, you can sign in later and claim the project.

# Agent: provision a claimable project (writes DATABASE_URL to .env)
neon claim create --env-pull

# Human: sign in and take ownership
neon claim accept

Today you can provision Lakebase Postgres, the Data API, and Managed Better Auth this way, with support for Object Storage, Functions, and AI Gateway coming soon.

For how it works, see the Claimable Neon docs and the launch blog post.

If you're building an AI agent or app that manages Neon itself, such as one that spins up a project or a branch, @neon/tools gives it typed, function-calling tools instead of hand-written API wrappers. It turns the @neon/sdk client into tools you hand to a model. Pick the Neon operations you want to expose, and they stay in sync with Neon's API. It also ships adapters for MCP, Mastra, and Eve.

import { createNeonTools } from '@neon/tools';

const tools = createNeonTools({
  apiKey: process.env.NEON_API_KEY,
  tools: ['projects.list', 'projects.createAndConnect', 'branches.createAndConnect'],
});

// createAndConnect provisions a project, attaches compute, and returns a connection string
const created = await tools['projects.createAndConnect'].execute({
  name: 'agent-project',
  region_id: 'aws-us-east-1',
});

The same @neon/tools catalog now powers the hosted Neon MCP server for coding assistants like Cursor and Claude, adding over 70 new tools. Read more in Give your agent Neon tools.

Lakebase Postgres now puts much more of a large compute's memory to work as cache. On fixed-size computes of 18 CU and above, shared buffers expand to use up to 75% of available memory, backed by huge pages, so more of your working set stays resident and fewer reads fall through to storage. Cache-heavy workloads can see up to 2x throughput, with no configuration change or restart required.

On an affected compute you can confirm the larger cache with SHOW shared_buffers and SHOW huge_pages.

For the engineering deep dive, see Improving Lakebase Postgres compute cache on Neon, Part 1.