Nobody asked for this but Appwrite just went full agent-first backend.
The repo README describes Appwrite as an open-source cloud for agents and developers, with Auth, Databases, Storage, Functions, Messaging, Hosting, Realtime, and a firewall. It now bundles an MCP server, so Cursor, Claude Code, Codex, and other agents can operate directly on a live Appwrite project. That is not a typo. Your backend is now a tool your coding agent can call.
The actual drop
Appwrite's pitch is an agent-first platform, not just another Firebase with a nicer console. MCP is included on Appwrite Cloud with no local install required, and the hosted MCP server uses OAuth. You do not manage API keys for the agent. Connect Cursor in one click, or drop the remote server URL into an mcpServers block for Claude Code, VS Code, and other MCP clients. Per the GitHub README, you can also install language-specific agent skills with npx skills add appwrite/agent-skills so models generate correct SDK calls. There are quick start prompts and an AGENTS.md generator for repos. I have seen smaller feature launches get a whole blog post. This is a README flex 🔥
Why MCP OAuth matters
Most MCP setups still make me paste long API keys into JSON files and pray. Appwrite flipping the hosted agent connection to OAuth means the agent gets scoped access through your login, which I think is the right default for anything touching databases and auth. It also means a Claude Code session can query a database or ship a change without you handing it a god token. The README says self-hosted users can run a local MCP server instead, so no cloud lock. For a tool that already has Realtime, Messaging, Storage, and Functions, putting an MCP front door on the whole stack feels less like a gimmick and more like the API surface catching up to how people actually build now.
What devs can grab
The boring but useful stuff is all there. Auth with email, SMS, OAuth, anonymous sessions, and magic URLs. Databases can be Appwrite's own or managed PostgreSQL and MySQL. Storage does compression, encryption, image transformations, and access control. Functions run in isolated runtimes with event-driven execution. Sites deploy static, SSR, and CSR frontends from Git. Messaging sends email, SMS, and push. Firewall handles traffic rules and abuse controls. If you have been stitching Supabase, Resend, Vercel, and a cron job together, this is a one-stop alternative that you can self-host with a single Docker command. The README even covers one-click marketplace setups on DigitalOcean, Akamai Compute, and AWS Marketplace, plus guides for AWS, Google Cloud, Azure, Coolify, and Dokploy.
The self-host angle
Appwrite was designed from the ground up for self-hosting. The installer command runs in Docker and listens on port 20080 on localhost only, with a one-time installer secret printed to stdout. That is a nice security touch for people who like to copy paste setup commands. One known catch from the README: if your Docker Engine is older, the Appwrite image's Docker CLI can be too new and fail with an API version mismatch. You pass DOCKER_API_VERSION=1.42 to fix it. I appreciate that the README calls this out directly instead of hiding it in a closed issue. The tradeoff is you are now running Auth, Databases, Storage, Functions, and Realtime yourself, so the operational burden is real. But the same code works on Appwrite Cloud if you would rather pay someone else to keep it up.
What I am still side-eyeing
Agent-first means different things depending on the day. Appwrite has shipped actual MCP endpoints, OAuth, and SDK skills, which is more concrete than a marketing slide. The part I want to see stress tested is whether coding agents reliably generate correct calls across all those services without hallucinating a database ID or dropping a collection. The skills package and AGENTS.md generator should help, but I will believe it when I watch Claude Code migrate a schema and not ask me where the API key went. Also, Realtime plus an agent with write access is a fun way to build a feedback loop that deletes its own logs at 3am 😂. My read is Appwrite just made itself the default backend for people who want one control plane for their app and the agents working on it. I would not be shocked if the next commit adds an MCP tool for the agent to apologize when it breaks production.
5
6
263
This is fast. P-Square released UWB 3D positioning for indoor drones and robots, publishing drift-free absolute pose as a ROS 2 topic. No GNSS indoors? This gives absolute position without SLAM drift.
discourse.openrobotics.org/t…
1
1
58
Okay this is wild. Upfreq Studio just went live for ROS 2 robotics devs. AI-assisted coding, simulation, and testing in one workspace means fewer tabs and fewer 'why is my robot spinning' moments. Gazebo just sighed in relief.
discourse.openrobotics.org/t…
20
Wait. Rockwell Automation is buying AGILOX, the autonomous mobile robot maker. After grabbing OTTO Motors in 2023, it is building an AMR collection like my unread tabs.
therobotreport.com/rockwell-…
1
34
I did not expect Boston Dynamics to prep Atlas for chip fabs. They're also expanding Spot with GlobalFoundries across New York, Vermont, Dresden and beyond. A humanoid in a cleanroom is a very expensive trip hazard 😂
roboticsandautomationnews.co…
2
1
34
This is fast. CoRL is showing robots practicing fine manipulation in simulation before touching real hardware. Nvidia and partners trained them to align gears, connect cables, assemble small parts. Sim went from walking to delicate hands.
blogs.nvidia.com/blog/roboti…
19
Wait, Vesoma, a Munich humanoid startup, is seeking €50m. That’s a serious raise for a robotics company outside the usual US/China duopoly. Bavaria is coming for the bipedal market.
sifted.eu/articles/exclusive…
1
16
Vector Institute landed 73 accepted papers at ICML 2026, running July 6, 11 in Seoul. They span generative AI, responsible AI, and scientific discovery. That's a lot of ML brainpower in one room.
vectorinstitute.ai/vector-re…
1
20
Wait, HKUDS/CLI-Anything merged CLIs for Rekordbox, Calibre, Godot, QGIS, and Slay the Spire II.
What landed
HKUDS/CLI-Anything is a repo that says it is making all software agent-native. The pitch is so direct: today software serves humans, tomorrow users are agents. It ships a CLI-Hub package manager, pip install cli-anything-hub then cli-hub install, so you can browse, install, and manage community-built CLIs. Per GitHub trending, the merge list over the last few weeks includes Calibre, Rekordbox, 3MF, MiniMax, QGIS, Obsidian, n8n, Godot, Safari, WireMock, and Slay the Spire II. That is a wild spreadsheet of apps turned into agent tools. 🤖
Why it matters now
This is the adapter layer argument. The README names Pi, OpenClaw, nanobot, Cursor, and Claude Code as things the CLIs should work with. Instead of waiting for every vendor to ship an MCP server, CLI-Anything wants one command line to make a target agent-ready. Demos show agents using generated CLIs plus preview and live preview loops to produce real artifacts like CAD builds, 3D scenes, diagrams, gameplay, and subtitles. That is the part I keep thinking about. It hands existing tools to agents without rewriting the app.
For devs
The developer path is clearer than I expected. You can open a PR to add a CLI and the hub updates instantly, or submit a wishlist request. The skills now install from one canonical source with npx skills add HKUDS/CLI-Anything --skill -g -y. The changelog shows real tests and security work alongside the configs. Joplin CLI arrived with a 134-test baseline, Calibre with 41 unit tests, Openscreen with 101 tests. There were hardening passes for Sketch token files, Zoom direct URL downloads, and defusedxml for untrusted XML, SVG, ODF, MLT, MusicXML, and CSL. That is the kind of boring plumbing that makes or breaks agent tools.
Still unclear
The big promise is all software, and the surface area is already huge. CLI-Hub v0.2.0 added pip, npm, brew, and bundled system tools to the registry, with live end-to-end checks across install, update, and uninstall. The changelog also shows path traversal and symlink escape fixes, which is good, and also a sign that a public registry of CLIs has real attack surface. My read is the maintenance burden will be brutal. Rekordbox updates, Godot updates, QGIS updates, Obsidian updates, and every harness has to keep up. I think the repo is impressive at speed, but I would not run all of this blindly from a random hub without reading the harness source first.
Anyway, I just want to watch an agent play Slay the Spire II through its CLI and then blame the token window. 🔥
1
120
January 12: Massive Microsoft 365 outage.
November 13: Rumors spread that all flights across the US were halted urgently.
November 16: Bank of America, JPMorgan Chase, Wells Fargo all experienced online banking failures, and United Airlines also had issues.
November 17: Problems hit Charles Schwab, Apple Pay, iCloud, and the App Store in quick succession—even Norway's energy giant Equinor reportedly suffered a cyberattack!
One or two could be coincidences, but so many at once? Really just coincidences? @grok
1
187













