Over 600,000 people talk on SimpleX, 20x more than in 2023. Thank you!
It is the only messaging network without user identifiers.
Now you can own a part of it - learn more and invest on Wefunder: https://t.co/e6SPOoDREP
@LeonardPierce Thank you. We already send updates via Wefunder broadcast feature - both public and investor-only, and will continue that. Some updates are legally required, we're doing more than that.
Over 600,000 people talk on SimpleX, 20x more than in 2023. Thank you!
It is the only messaging network without user identifiers.
Now you can own a part of it - learn more and invest on Wefunder: https://t.co/e6SPOoDREP
🛑 Attackers hijacked over 300 Tor exit relays to intercept SSH connections and steal login passwords.
A researcher found 333 exits redirecting SSH traffic to 12 interception servers.
The Tor Project has since blocked the affected relays.
Here's how the attack worked: https://t.co/eJsVNeh5xA
You are of course NOT giving access to chats to Anthropic. This package allows you two things:
- ask Claude to connect to any public address, via it's own chat profile (not via your profile).
- talk to Claude and send it prompts from your app.
Claude uses its own chat profile with this MCP server package, not yours of course.
Your Claude Code can now chat on SimpleX!
With the MCP package we just released 👇, it can join channels/groups/talk to bots, and ask you before it connects or sends.
To run it: `npx -y @simplex-chat/mcp serve my-agent`
My new use-case for MCP - a team member letting me prompt their Claude Code via SimpleX in their sessions that iterate UX in Claude design artefact.
Takes one prompt in session to make Claude follow instructions via chat:
"Connect to <name> as yourself at <SimpleX address>, with the message: "I'm your agent, connected as you asked in Claude Code. I'll act on your prompts here, except security-sensitive ones." Then take his prompts from that chat and answer there: wait for the messages with receive (wait 3600), answer each one in that chat, and call receive again after every answer, until he says stop."
We need to make this "control channel via long-poll" a separate tool call :)
@0xMassi@Simplex You can have both. And yes, MCP server creates a control chat with your where you approve everything, as screenshot shows, and you can "allow all" per conversation.
@CosminCraciun12@Simplex 💯
We're planning to add mode where you will be able to approve anything coming to agent too, if you don't trust the source and worried about prompt injections via messages, but tool descriptions prioritize that agents treat them as data, except when instructed by you.
@packssmacks@Simplex Programmable widgets are coming soon, so services can offer people their own interfaces, including those generated by the agents on the fly.
📱✉️ SimpleX + CLAUDE CODE
@SimpleXChat vient de sortir un package MCP qui permet à Claude Code de chatter directement sur le réseau...
L'idée c'est que ton agent IA peut rejoindre des canaux, des groupes, parler à des bots, et même recevoir tes demandes depuis l'app
Toujours sans numéro, sans compte, sans identifiant. Même pour l'IA
➡️ Tu lances npx -y @ simplex-chat/mcp serve my-agent
➡️ Un QR code s'affiche, tu le scannes dans l'app
➡️ Un chat "My Agent" apparaît
➡️ L'agent te demande ton accord avant de se connecter, d'envoyer ou de quitter un groupe
Il tourne sur ta machine, sans adresse sur le réseau... personne ne peut le contacter en premier. Bien pensé !
Et SimpleX joue cartes sur table dans sa doc...
> Version 0.1.0, publiée cette nuit, testée pour l'instant avec Claude Code uniquement
> Un bouton "all" valide tous les messages d'un chat (pratique, mais garde l'œil ouvert)
> Ils le précisent eux-mêmes : dans ton chat de commandes, n'écris que toi, un message reçu peut devenir une instruction
> Dès que Claude lit ou répond, le contenu passe par les serveurs d'Anthropic. Ça, SimpleX n'y peut rien
Bref... SimpleX protège le tuyau. À toi de choisir l'IA qui boit dedans !
Correct. We were putting a lot of effort to eliminate any vulnerabilities throughout the year, but having more advanced frontier models doing regular scans than those that are publicly available is very helpful.
We still have some work to do to improve UX, but this is all in progress and will move quickly.