8 projects

Ask your AI assistant for a podcast episode about anything you like. The assistant plays editor-in-chief: it finds the sources, asks the studio for a draft over MCP, and goes through the script with you. The studio is a Symfony app: a writer agent turns the brief into a dialogue between two AI hosts, each line is voiced to match its emotion, and the episode lands in a real podcast feed with chapters and a transcript. Listeners can also ask the hosts a question by voice, and the hosts remember past episodes. Built on the MCP Bundle and PHP MCP SDK, an AI Bundle agent, structured output, Voxtral for speaking and listening, and embeddings in Postgres. Two pull requests to symfony/ai came out of it, #2873 and #2894. It's deliberately a showcase: the same building blocks could just as well power interactive stories, release podcasts or language lessons.

FormBridge lets an AI assistant fill out the forms your Symfony application already uses. Add one PHP attribute to an existing form handler, and FormBridge exposes the form as an MCP tool. It builds the tool’s input schema from the form’s fields, choices, and validation constraints, then returns Symfony’s own validation messages so you only maintain one set of rules. In clients that support MCP Apps, the person sees the prefilled form right in the conversation. Symfony renders it with Twig and shows each error next to the field it belongs to. The assistant fills in the details, and the person reviews, edits, and submits the form. The application’s existing handler saves it as usual. Other clients get a link to open the same form in a browser.

NextStep MCP turns any Symfony Workflow or state machine into an MCP server where an AI agent only sees the tools that are valid right now. Each transition in your workflow becomes an MCP tool automatically, and when the state changes the server tells the client its tool list has changed: once an order moves from paid to shipped, “refund” disappears and “request return” appears. Workflow guards stay in PHP, so the agent cannot skip a step however it is prompted, and a call that is not valid for the current state is refused with the list of transitions that are. Transitions that need a human detail, such as a refund reason or a confirmation, ask for it through MCP elicitation, and the current state and a live diagram of the workflow are exposed as an MCP resource. Built as a reusable bundle on symfony/workflow, symfony/mcp-bundle and the official PHP MCP SDK, with a demo order workflow that runs from a clone in one command and needs no API key.

Any agent (Claude Code, Cursor, VS Code, Zed) runs shell commands through one MCP server. Each command is checked before it runs: - Asked in your own client. A risky command returns an MCP elicitation form, which you answer with once, this session, always for this project or refuse. The MCP bundle signs the request state, and Airlock ties it to that exact command with a one-use token. An answer for `make test` can't be reused to run anything else. - Read like a shell would. A grant for `make test` doesn't cover `make test; rm -rf ~`. `sudo`, recursive deletes, `git push` and `curl | sh` are always asked. - Run in a sandbox (bubblewrap on Linux, sandbox-exec on macOS). Commands can write only inside the project and can't read your keys. The network reaches package registries and the hosts you allow. - Output checked by a model through `symfony/ai-platform`, using structured output (`clean` or `injection`). Secrets in the output are masked, and every command goes into one audit log. Built on `symfony/mcp-bundle` (`mcp/sdk`), `symfony/ai-platform` and `ai-bundle`. It installs with one command and has a working demo in five minutes.

FlyNotes is a collaborative game where humans and AI agents exchange messages through virtual paper planes. Symfony AI creatively reinterprets each user’s message based on their emoji, while Sky Pulse, an internal AI agent, uses tools to carry out missions and publish results as new planes. Human-in-the-Loop workflows let users approve reinterpretations, authorize tools, and confirm results before publication. Through MCP, external AI agents can also discover messages, launch planes, and leave stamps. Try it yourself! 👉 https://flynotes.vitaminatech.com ✈️

When an AI agent can refund, email or delete, its rules can’t live in its prompt: a persuasive user talks the model out of them. Tool Guard puts the rules in code, on the tools themselves, with the same PHP attributes for Symfony AI agent tools and for MCP tools: #[Granted] (Symfony Security voters, with a tool argument as subject — on MCP tools too, beyond per-tool OAuth scopes), #[Limit] (numeric ceilings: deny, or hold for a human above a threshold), #[RateLimit] and #[RequiresApproval]. Each agent also has its own role: a call runs only if the user and the agent are allowed, so the boss’s support assistant still cannot export every customer. Below the limits the agent acts alone. A refused call never runs and the model is told why in plain words; a held call waits for the app owner, who is told through Symfony Notifier (Slack, Teams, Telegram, email, SMS) or by a phone push, and approves or rejects the exact arguments, never text written by the model. Every call and decision goes to a hash-chained journal, so a deleted attempt breaks the chain. Works with symfony/ai-bundle, symfony/mcp-bundle and the PHP MCP SDK. The demo runs in one command with a scripted hostile agent, no API key needed, and the same attack can be played against a real model (gpt-oss-120b on Hugging Face).

First time participating in a Hackaton, so project will be used as a Hackaton template.

A code review interface for developers who accept AI can be useful, yet still read the code. After week's worth of AI work, there is a lot to review, even if you provided guidance at every step. Slopmower lets you ask questions, call out slop, request changes and provide guidance to your agent. Slop keeps growing, so it needs mowing.