OCTOBER 2026

Workbooks in the pganalyze MCP Server, the Postgres 19 webinar replay, Webhooks early access + PGConf.EU next week

Hi there,

It's been a busy few weeks in Postgres land! With a slight delay, Postgres 19 RC1 came out this week; RC2 is now scheduled to land with the November 12 patch releases and the final release is planned for November 19. We used the occasion for our Postgres 19 webinar. 19 has many great improvements - pg_plan_advice is a personal favorite that enables query plan control (like pg_hint_plan, but better at dealing with complex queries), followed by REPACK CONCURRENTLY, and also brings some notable default settings changes: JIT is now off by default, and default_toast_compression changes from pglz to lz4. If you missed the live session, the recording is now available.

At pganalyze, we've been pushing forward how we can safely use AI with production databases: Connecting an AI coding tool directly to production is usually a bad idea: you don’t know what it will run or what load it adds. For query optimization that leaves the feedback loop broken: Claude Code, Codex or Cursor can read the statistics and plans pganalyze already collects and propose a rewrite, but the tools usually can't run EXPLAIN ANALYZE.

Benchmarking with Workbooks in the pganalyze MCP Server, now in public preview, closes that loop: the agent creates a Workbook for the slow query, pganalyze runs EXPLAIN ANALYZE through the collector, records the baseline, and measures every rewrite as a variant against it. The agent never gets a database connection. The collector executes EXPLAIN ANALYZE and nothing else, in a read-only transaction that aborts after 60 seconds, behind a separate write scope and the same Create Workbooks permission the app requires. You confirm the parameter values to ensure reproducability, and the agent leaves notes on the Workbooks with what was attempted for later review. The announcement post walks through how it works end to end, including a live demo by Ryan Booz.

There's also a new way to get Alerts & Check-Up issues out of pganalyze and into your own tooling: Webhooks, now in early access. pganalyze sends each issue as a signed HTTP request to an endpoint you control, following the Standard Webhooks specification, so you can verify requests with any Standard Webhooks library and route them into whatever runs your on-call. If you'd like to try it, let us know by replying to this email, and we'll be in touch.

And on the community side, PGConf.EU starts next week in Valencia, October 20 to 23, and we're there as a Platinum Sponsor. Ryan speaks on Tuesday about Postgres tuning in the age of AI, and why context is everything. On Wednesday I (Lukas) will give a sponsor keynote on how the way we tune Postgres is evolving, and on Thursday we'll talk about the future of plan management and feedback loops. On Friday morning, the PostgreSQL Observability Summit I'm co-organizing wraps up the week. If you're in Valencia, come say hi! After that, you'll find us at KubeCon North America in Salt Lake City and at AWS re:Invent in Las Vegas; details are in the events section below.

As always, you can find the complete list of pganalyze product updates, including a collector release with a security fix, at the end of this newsletter.

Announcement

Safely enable AI Coding Tools to use EXPLAIN ANALYZE on production: Introducing Workbooks in the pganalyze MCP Server

AI coding tools shouldn’t get a connection to your production database, but that leaves their feedback loop broken: Claude Code, Codex or Cursor can propose a query rewrite from the statistics and plans in pganalyze, but can’t run the EXPLAIN ANALYZE that verifies whether it worked.

Giving them direct access wouldn't be safe - but what if we can design purpose-built guard rails? With the Collector Workflow for pganalyze Workbooks, we've previously launched a safe way to run EXPLAIN ANALYZE against production servers or replicas/clones with production data. Workbooks benchmarking in the pganalyze MCP Server, now in public preview, enables AI tools to benefit from that as well.

The agent creates a Workbook for the slow query, pganalyze runs EXPLAIN ANALYZE through the collector and records the baseline, and every rewrite or planner setting change becomes a measured variant with its thesis recorded in a shared Workbook. Ryan Booz and Keiko Oda walk through how it works end to end and what else changed in the pganalyze MCP Server since its initial release in April.

Webinar Replay

Postgres 19 in practice: what shipped, what got cut, and how to test it

Recorded when Postgres 19 RC1 was released (yesterday!), Lukas walks through what (most likely) ships in 19 and how it can make real-life production situations easier: REPACK CONCURRENTLY and the background on what kind of bloat it removes, WAIT FOR LSN for read-your-writes on replicas, pg_plan_advice for controlling query plans and pg_stash_advice for pinning them, how pg_stat_plans 2.1.1 integrates with pg_plan_advice to show advice tags for plan statistics, new monitoring views such as pg_stat_lock, and the EXPLAIN ANALYZE timing changes he contributed together with others.

Lukas closes with the changed defaults (JIT off, lz4 TOAST compression), sequence replication and LISTEN/NOTIFY scalability, the release schedule (RC2 on November 12, final release on November 19), and how to test the release candidate and report bugs.

Community News

Upcoming Events

PGConf.EU 2026, Valencia, October 20 to 23. pganalyze is a Platinum Sponsor, with three talks from the team: Ryan Booz, "Postgres Tuning in the Age of AI: Context Is Everything" (Tuesday 13:45, Audit 3); Lukas Fittl, "Evolving How We Tune Postgres" (Wednesday 16:55, Audit 1, Platinum Sponsor Keynote); and Lukas with "The Future of Plan Management and Feedback Loops" (Thursday 11:25, Audit 3). On Friday morning, Lukas co-organizes the PostgreSQL Observability Summit, a half-day community event (09:00 to 12:30, Audit 2). If you're in Valencia, come say hi!

KubeCon + CloudNativeCon North America, Salt Lake City, November 9 to 12. Come find us at booth 324 in Hall 4 and talk through finding slow queries across your clusters, and learn how CNPG integrates with pganalyze.

AWS re:Invent 2026, Las Vegas, November 30 to December 4. pganalyze is exhibiting again this year: find us at booth 1750. If you run Postgres on Amazon RDS or Amazon Aurora, come by and see how pganalyze finds the queries that cost you the most.

Product Updates

What's New in pganalyze

Alerts & Check-Up

• Webhooks (early access): send Alerts & Check-Up issues as signed HTTP requests to your own endpoints, following the Standard Webhooks specification. Add a webhook under Settings > Integrations, then route checks to it through Alert Destinations. Delivery history, retries and secret rotation are built in. Reach out to join the early access.

pganalyze MCP Server

• Workbooks benchmarking is now in public preview: eight workbook tools (create_workbook, create_workbook_variant, create_workbook_parameter_set, update_workbook, get_workbook_details, sync_workbook, get_insight_rewrite, delete_workbook_item) for any MCP client with the write_workbooks scope and the Create Workbooks permission
• New get_system_stats tool: OS-level time series for CPU, memory, storage, network and WAL size
• Support for connecting from Devin
• Personal access tokens (early access): a scoped credential for scripts and background agents that acts as a specific user

Queries & Connections

• Connection Traces now link to autovacuum runs in progress, instead of showing them as an unparsable query
• Fixed the EXPLAIN plan list timing out on queries with many captured plans

Other improvements

• Icon-only buttons are now named for screen readers
• Color contrast has been improved for better accessibility

pganalyze-collector 0.75.0

• Update pg_query_go to 6.2.5, which includes a security fix: certain crafted utility statements (for example DO ... LANGUAGE, or CREATE/ALTER SUBSCRIPTION ... CONNECTION) could cause the collector to leak process memory into normalized query text, or crash (GHSA-6ggm-xmc9-8ffg); it also adds normalization support for NOTIFY statements
• Azure: support db_use_iam_auth for Azure Database for PostgreSQL, connecting with a Microsoft Entra ID access token fetched with the same credentials already used for Azure Event Hub log collection (client secret, client certificate, workload identity or managed identity)
• Crunchy Bridge: detect the cluster ID from internal hostnames (i.*.db.postgresbridge.com)
• Log Insights: correctly classify non-aggressive anti-wraparound autovacuum log events ("automatic vacuum to prevent wraparound of table")
• Add schema version tracking to snapshots