Hello, I’m too lazy to explain at length. Basically, this is MCP Server that allows Agent A to chat with Agent B in our Cursor IDE. In particular, you can also chat directly with both Agents to assign tasks for them to perform, or adjust the task when you see that the two Agents are deviating from the original plan.
Luckily for us, Cursor currently does not limit the number of toolcalls, so this MCP Server works perfectly while still SUPER SAVING RATE LIMIT.
I have used Claude 4 sonnet MAX + Gemini 2.5 Pro for many consecutive days without hitting the rate limit. Maybe because I haven’t used it enough, but if I don’t use MAX mode, I SURE to you that I will NEVER hit the rate limit.
This tool is free. Thanks for using it.#1
Important note: people who have used MCP Server AI-interaction before will easily use this tool without any difficulty.
@condor I used Claude 4 sonnet MAX + Gemini 2.5 Pro to write Rule_For_AI for this tool. It’s really awesome. What do you think if Claude 4 Opus works with Gemini 2.5 Pro MAX? 1 model in charge of the main code and 1 model in charge of storing the giant context window of up to 1 million tokens.
Claude 4 Opus + Claude 4 Sonnet? Claude 4 Sonnet + O3 Pro? Really unimaginable!
if i understand this correctly agent 1 can be gemini with 1mil context and agent 2 sonnet 4, gemini will receive messages from me, search codebase and store relevant data and provide it to sonnet 4, which completes the task and returns it to gemini to which I can further communicate another tasks while preserving previous context? i am not sure there is a large potential even if this works as cursor handles context rather well in my opinion, but mainly i think the project needs a little better description and showcase of what it actually does, meaning its purpose and advantages, preferably a video, what do you think?
Thanks for your interest. I have a user guide in Readme.md of github, please visit and read it. I tried to make this tool as easy to use as possible. Try it out. You won’t believe what 2 Agents will say to each other, it’s really interesting.
Warning: don’t use MAX MODE if you don’t have enough money because activating 2 MAX MODE Agents at the same time will burn your rate limit very quickly. If you don’t use MAX MODE, you can use it all day, all month comfortably.
About the tool you sent me, I took a quick look at it and found it to be a useful tool. There is also a web client, I see that some tools like mine also create web clients, desktop clients, quite convenient right? But for me, I won’t create more clients, because I don’t want to spend more time designing them. I created AI-interaction and Agent-to-Agent for my own use, so I just need it to be fast, convenient, and effective. I post the article here to share with those who want to use this kind of tool, and can edit the source code yourself to add the features you like.
Thanks for your interest. I simply created a tool for 2 Agents to communicate with each other and also communicate with the user (you). The purpose is to improve the quality of the output, 2 Agents working together to solve a problem will create a truly better quality result than just 1 Agent. If you are someone who knows how to take advantage of the strengths of each Agent, you can assign roles to the Agent, for example “Hey Claude 4 sonnet, you will be the one to edit the main file. And Gemini 2.5 Pro, you will be the one in charge of supporting and cooperating with Claude to perform this task”. They work quite well together lol.
In addition, you can chat with just 1 Agent and let Agent 2 join whenever you want. I rarely use this tool because I only use it when I want to develop a cool library, for example: virtual table with drag-and-drop columns and rows, merge cells, copy cells, paste cells, etc.
For instructions, please read Readme.md in the Github link. Thank you.
Thanks for sharing this and your attention. Agent-to-agent seems cool project. Cursor Team did 3 agents in 1 you probably saw that new update. But I prefer to use Claude 4 and Gemini 2.5 pro 06-05 preview so I don’t need to use o3. I think claude and gemini is better than gpt-4.5 and o3.
Is this generating new Agent S+ Modal? There is too many things on screenshot. I didn’t try it by myself but I’m just trying to understand the UI. I don’t need to use MAX Mode for Large Context, non-MAX mode is good for me. But now like everyone else I’m stuck on Rate limits. I want to use that tool and maybe I can improve that. I read readme.md file and I can say I’m still confused. Can you tell me Admin chat UI and controller UI and tab agent how they are works where we going to input prompts and getting answers from AI?
Besides, I can see we’re loading a user rules to cursor for that software, But I’m using my own User Rule too this seems can cost large context am i wrong?
Can you record a simple video for using this tool please? Because I want to use this software at best way like you.
Agent S+ Modal is just a Custom Mode that I created and named, to be able to create a Custom Chat Mode yourself, you can go to Cursot settings, it has an on/off button.
About user rules. I tried to use Claude 4 sonnet and Gemini 2.5 Pro to design it so that it is complete but still concise, easy to understand, easy to remember, easy to maintain for long term for Agents. You can add it to your user rules section, and use Claude 4 sonnet + Gemini of this tool itself to support you to rewrite user rules so that it is most effective.
Do you have facebook? I will send you a video tutorial if you still do not understand the explanation below.
Let me explain, it’s easy:
Step 1: I tell Claude 4 sonnet “Hey bro, start Agent 1 now”
Step 2: I tell Gemini 2.5 pro “Hey bro, start Agent 2” note → with the user rule I’ve set up, Claude and Gemini will easily understand that they will have to call tool agent chat 1 (Claude) and agent chat 2 (Gemini)
Step 3: I see Claude and Gemini call tool agent chat 1 and agent chat 2 → I click the “run” button for both of these tools
Step 4: I will look at the Controller UI and see messages from both Agents in the “Message Queue”. And in Waiting Agent I also see Claude and Gemini are in the status of waiting to receive messages
Step 5: I read Claude and Gemini’s messages
Step 6: I want Claude to be the first to greet, so I will delete Gemini’s message in Message Queue and click on the “Smart Delivery” button, this button will automatically search for the Agent waiting for messages in Waiting Agent to send messages to that Agent (I have processed so that this function does not send the sender’s message to the sender himself). In this case, Claude’s greeting message will be sent to Gemini → Claude is still in the waiting message state because Agent chat 1 has not responded yet → Gemini receives Claude’s message in the response section of Agent chat 2 tool → Gemini calls back Agent chat 2 to reply to Claude → I see Gemini’s message on Controller UI → I click the “Smart Delivery” button → Gemini’s message is sent to Claude → Claude receives Gemini’s message in the response section of Agent chat 1 tool → Claude calls back Agent chat 1 to reply to Gemini → I see Claude’s message in Controller UI → I continue to click the “Smart Delivery” button → the loop continues
Step 7: I want to assign tasks to these 2 agents → I click the “AI chat” button in Controller UI → “Admin Chat UI” is displayed → I enter a message and click the “Send” button → all Agents in Waiting Agent will receive my message → Both Agents calling Agent chat 1 and Agent chat 2 will receive a response, that’s it that’s my message (Admin) sent → Claude and Gemini will discuss and divide the work to do the task.
If you have any difficulties, please ask me, I’m very happy that someone wants to use my tool. Let’s develop the source code to create a great tool for you.
except that, when i start start ai agent and etc. cursor and modals giving “your account is suspicious contact to …” error. So you can think about to change that commands
Its flagging the mcp or something but I can’t use gemini 2.5 pro or claude 3.7 sonnet rn because of that error. I used that tool and it did happen to make because of start ai_interaction start agent chat 1 commands
Yes my friend this tool flagged by cursor I guess It’s bad you should check it with Pro Trial account. I was testing with my Pro Trial account because I don’t want to spend my Pro plan limits so that happens. And now my pro trial account just terminated I’m Free plan on this account.
Looks like you’re blocked by Cursor. It’s not my tool, it’s your account. Maybe free Cursor has its own limitations. Are you afraid my tool is malicious? Lol. Why don’t you use your real account?
I don’t need help here my friend I just gave feedback. I can’t use this tool because its really complicated and flagging I guess i don’t know maybe I sent too many request.
Maybe modal agent blocking or ignoring the “start ai agent” commands. Try with Pro Trial and check it if its good for you maybe i did something wrong have a nice day
I sent many request quickly with 2 tab like stopping and sending start ai agent etc. I’m sad about i can’t use this tool It could have helped me…
It could have helped me…