NeuroREPL.jl - Community Interest Check

Tooling
,
1

Hey, NeuroREPL.jl is a Julia implementation of the Model Context Protocol (MCP) that transforms the Julia REPL into a programmatic API for AI agents. NeuroREPL provides process‑isolated code execution, dual MCP transport support, and introspection tools specifically designed for agent workflows.

The system is built on Malt.jl for worker process isolation, HTTP.jl for dual transport (SSE and Streamable HTTP), and provides 28 specialized tools across 8 functional categories including core execution, package management, code introspection, performance analysis, and server control. The architecture separates the MCP server from the execution environment, ensuring that code execution cannot crash the server while maintaining persistent state across tool calls.

NeuroREPL.jl is inspired by the MIT‑licensed MCPRepl.jl project but represents a significant rewrite focused on improved agent communication, and enhanced debugging capabilities. The implementation is contained in a single ~2000 line Julia module for ease of modification and understanding. Initially my idea was to use Distributed.jl, however, a colleague of mine (I hope he does not mind) suggested Malt.jl and I think it could be a right choice (thanks RA).

Testing has been performed with Kiro CLI v1.25.0 (AWS), Crush v0.43.0 (Charm), and OpenCode v1.2.10. While core functionality is rather solid, the package remains in active development and is not quite production ready.

Some current limitations include:

  • No LSP Integration
  • No Debugger Integration
  • No Full REPL Features in Worker Mode
  • No Binary/Image Output Support
  • Limited Error Recovery
  • Single Worker Process for MCP Requests
  • No Explicit Session Management via MCP Protocol
  • No Security Layer

Potential enhancements include:

  • Parallel Agent Collaboration
  • Multi-Model Solution Validation and Ranking
  • Meta-AI System: Competitive + Collaborative Multi-Model Approach

In case there is any interest from the Community, I would be happy to publish the code. I was thinking about using the MIT License, however, in case of anything more substantial and/or some serious ideas, of course, I remain open to suggestions.

A screenshot showing the starting sequence and functionalities is attached below.

s1a_s2b
s1a_s2b921×1895 146 KB

3 Likes
2

I’d so definitely download and set it up on a LAN server.

1 Like
3

Great. I use a mesh virtual private network, so everything works like a LAN. Anyway, localhost is the default setting, so you will probably have to update the IP address. I just uploaded a draft of the documentation. It is available here: NeuroREPL.jl/docs/NeuroREPL.pdf at main · justcodebase/NeuroREPL.jl · GitHub. Do you have any suggestions regarding the license?

1 Like
4

I use MIT for everything—it’s minimalist, been around forever without any high-profile IP disputes and passes my lawyerly sniff test.

1 Like
5

I’m not a lawyer, however, I must admit that my professional experience began with research on competition policy and antitrust law. Anyway, I’ll think about it. If no one weighs in on this topic in this thread in the near future, I’ll choose the MIT license. Still, I have to admit that I have a feeling that it could potentially lead to something more substantial. I hope you do not mind this brief delay.

6

There’s a tension between open source and commercial protection of intellectual property. Ubuntu’s model is that it offers enhanced versions and services. You should get professional (paid) advice if you want to preserve your options.

7

I’m afraid that retainer fee based legal advisory falls outside the scope of our current discussion, at least for now. However, a capital raising style structure based on a success fee potentially using a lower spectrum of single digit percentages, if applicable, can certainly be discussed.

Re: Ubuntu. In the medium to long term, I expect Ubuntu to follow the Red Hat – IBM acquisition model, albeit at a valuation roughly 10 to 30 times lower than the one Red Hat exercised in 2019. That said, I prefer not to comment on its business model or financial results, I’m currently running Fedora 43.

Re: NeuroREPL. Anyway, did you have a chance to review the preliminary documentation? I don’t see much interest in this thread, however, I think the package is pretty good, so if it interests you, I’d be more then happy to grant you a personal license.

1 Like
8

Thanks. I’ll get back to you as soon as I’ve got my LAN server fully provisioned. (Amazing how much snappier the little i7 is without all the GUI.)

1 Like
9

Absolutely, when you’re ready, please send me a ping.

10

Just finished the documentation. Color me impressed. I’ve got kaimon and julia-mcp installed on my little lan ubuntu and am in the process of moving over serena, so NeuroREPL.jl would make quite a quartet.

11

Hey, I can’t comment on julia‑mcp because I haven’t tried it. My understanding is that it’s intended for local development due to the use of uv. I often switch machines, so rather it isn’t suitable for me.

Regarding Kaimon, I’d prefer not to comment. I’ve been in brief contact with its developer a few times, even invited him for a virtual beer and bought two Coronas, which turned out to be a total disaster. I tried again recently, my understanding is that it is some kind of a systemic problem, academic industry approach, perhaps. I have a meeting with legal advisors scheduled for the next week about my recent contact with the developer, and I’ve taken some notes.

This is the first time I’ve heard of Serena. It reminds me of Crush, which also supports the Julia LSP, though it looks a bit different. My experience includes Cursor, Windsurf, Trae, and Kiro on the IDE side, and Crush, OpenCode, and Kiro on the CLI side, plus brief encounters with Gemini and Rovo Dev. Currently I prefer Kiro IDE for initial package development, and Kiro‑CLI when I want to implement existing architecture into a new but similar package. I usually use Crush together with both Kiro IDE and Kiro‑CLI because it provides access to various models without context switching. Please take these preferences with a grain of salt. I’ve used all this software for over 1.5–2 years, and the field of coding agents is evolving rapidly. Do you think it’s worth trying Serena, if I may ask?

I’d be happy to provide you with NeuroREPL for testing. How would you like the code delivered? I can send the file via Julia Discourse Direct Message or by e‑mail (my address is on the first page of the documentation). I should note that, given the limited interest in this thread, I’m not currently interested in the MIT licence. Apart to making new friends, honestly, I was hoping for interest in discussing and implementing potential enhancements, especially regarding a Meta‑AI System - Competitive + Collaborative Multi‑Model Approach or its derivatives, which I mentioned in my initial post. I was considering starting with the Mamba 2 architecture, though it isn’t an easy project.

Are you guys here into startups at all?

1 Like
12

I crashed and burned a dotcom before it was fashionable, which kind of put me off startups. I might not be much use to you as a tester except as a naive user. My experience with MCPs is limited to the past few weeks.

13

That’s a valuable experience. More or less, at that time I got interested in this topic by one of an early Apple investors and shortly later I was gaining financial experience from early venture capital and private equity investors in my region on an advisory side, as well as, in a sense, almost from all leading global companies operating in the information technology space at that time. Apart from research, financial and technology sectors, I have an entrepreneurship background.

The major features of NeuroREPL that you may like:

i) You and your agent operate in an environment with only one very light dependency, so it is almost a clean environment.

ii) You connect over a TCP network stack, so you can have the agent on one machine and the code running on the other.

iii) You can simultaneously run multiple Julia instances connected to separate agents, you just need to choose a different port on both sides.

iv) You can connect multiple clients (agents) to the same Julia instance

v) It is relatively light (I was thinking about making it even lighter by replacing HTTP.jl with Libwebsockets.jl or JSON.jl with YYJSON.jl, but the complexity of the code would increase significantly)

vi) You can restart your Julia agent working environment very quickly

vii) There is support for custom logging, for me it was very useful to diagnose connection problems and compatibility issues with various agents

viii) There is quite extensive server‑status information available

ix) I believe the user interface is pretty cool, you can, for example, exit Julia from both server and worker mode with just :q!. Also the Julia prompt changes colors: from a native Julia green to Julia purple signaling that you are in server mode, and to Julia red signaling that you are in worker mode. I have also included support for color‑blindness and monochrome monitors.

As for your remarks related to startups, I have to admit that you made me think a bit about them today. The odds are like “one or two in a hundred” or maybe slightly better those days due to a much more mature ecosystem, but still not much higher because of the economics and investment cycle of venture capital funds. However, what is interesting:

i) those ideas that 30 years ago were a niche alternative assets investment class have, over time, propagated to the mainstream of active public markets asset management strategies, for example vide research papers by Mr. Hendrik Bessembinder on macro‑growth theory and empirical finance;

and

ii) to a significant degree, I believe, are associated with the 2025 Nobel Prize in Economic Sciences awarded to Mr. Joel Mokyr, Mr. Philippe Aghion, and Mr. Peter Howitt, which was about innovation driven economic growth, the process of creative destruction, and incentives for innovation.

Taking into account your Ubuntu example from the beginning of our conversation, despite all its shortcomings, I am still in favor of a classic startup model.

But hey, I am not pushing the idea of a startup. Recently I was briefly in touch with a person related to MCPRepl, the original package that got me interested in the MCP topic, and upgraded NeuroREPL to version 0.2. On the way, I spotted this idea related to a “Meta‑AI System,” which potentially might exhibit some startup characteristics. However, gaining a second mover advantage in this field will be hard. I guess it is always worth asking. In general, I remain interested in startup ideas, particularly in highly intellectually intensive areas, as this was one of the main reasons I joined the Julia community.

I will be off from this discussion list for the next couple of days. In case of interest in NeuroREPL, please drop me a line via e‑mail. I will reply as soon as possible.

1 Like
14

It’s not appropriate to talk about or threaten legal action against other community members here. There are three possible cases:

  • If there is a legitimate legal issue, and it involves the Julia community, then Stewards should informed and it should not be discussed in public Julia forums;

  • If there is a legitimate legal issue, and it does not involve the Julia community, then it should also not be discussed in public Julia forums and should be dealt with by relevant legal authorities outside of the community;

  • If there is no legitimate legal issue, then this is simply harassment of a Julia community member and may be cause for banning.

I don’t know the context and cannot determine which of these three cases this situation falls into, but in none of these cases is it appropriate to casually post about potential legal action against another community member.

16 Likes
16

Honestly, I think you are deeply misunderstanding the situation. I was not threatening legal action. What I wrote was that I was planning a meeting with legal advisors regarding the situation that took place on this forum.

I mentioned this as an honest response to some of the questions raised by one of the community members. The original intention of the thread I started was not to make comparisons with other packages, but simply to ask whether there is any community interest in the ideas outlined in the first post. Even though some of the topics discussed in the thread went beyond the scope of that original post, I did not want the interlocutor to feel that I was being impolite or uninterested in the discussion. In fact, I enjoyed the conversation.

In addition, actually a few weeks ago I contacted the moderators, politely expressing my lack of understanding of the situation. From my perspective, however, this unfortunately seems to have made things even worse. In general, after about five years here, I have to admit that I think that the level of legal and political correctness in the community is, to be honest, somewhat average.

In any case, I do not know whether there is a legitimate legal issue or not, these matters are not always entirely clear, not always straightforward and obvious, however, what I can say is that I am not worried.

I like the language and most people here are really great. I have always acted here in good faith. However, if you feel that banning my profile is necessary, I do not mind. In fact, I have already been significantly limiting my presence here recently.

P.S.

By the way, if you do not know the context and cannot fully determine the situation, there is probably no need to raise the voice, as it usually only makes things worse.