Show HN: Codemcp – Claude Code for Claude Pro subscribers – ditch API bills
github.comHi all! I normally work on the PyTorch project but I've been on baby leave for the past month, so I've been playing around with AI as a user rather than a framework implementor. I really liked the agent experience with Claude Code, but I couldn't really justify spending so many dollars on API costs for random side projects. I already pay for a Claude Pro subscription though, and it turns out you can simulate many of Claude Code's features with an MCP. If you have a Pro subscription, check this out! I think it really captures the Claude Code experience quite well, without forcing you to pay for API tokens.
I was wondering if I could use Claude Desktop to control VS Code directly (perhaps a plugin in it) so that I only need to open two windows.
This requires an MCP-compliant client, for which Claude Desktop seems to be the main (or only?) choice.
Note that Claude Desktop can be run on Linux using https://github.com/aaddrick/claude-desktop-debian, which repackages the Windows version to run on Linux (it's an Electron app, so it just needs to stub out the native interface, which seems mostly for cosmetic things).
It would be really nice though if the web versions of Claude, ChatGPT, etc. added MCP support directly: this should be achievable with help from a WebExtension and a native binary to proxy messages from the WebExtension to the MCP server.
It should also be possible to write such a WebExtension by a third-party (that injects messages in the conversation), although it's possible the LLM companies might block it.
I can report that I tried this setup (Claude Desktop on Linux + codemcp) on the Rust uutils repo asking it to add a "--repeat" option to cat to repeat the output N times (without telling it anything else), and it has generated a commit that looks plausible, correctly locating the implementation file and unit test file and changing them in an apparently correct way (other than mangling a commit just before the code changing \\n to \n).
It did require to manually enter "continue" in Claude's chat (and to approve the use of codemcp at the start), but it otherwise did everything automatically.
It seems to work.
codemcp automatically produces and commits a single git commit which also contains commit hashes for a bunch of other commits that contain a subset of the changes.
I just made an open source CLI based chat client that supports MCP servers to solve this problem yesterday at https://github.com/Flux159/mcp-chat.
The issue I faced during MCP server development of my Kubernetes MCP server (https://github.com/Flux159/mcp-server-kubernetes) is that I wanted to be able to easily test changes to the server without having to deal with Claude Desktop. Realized that I could make a simple CLI client that lets me automate some of the testing & it just happened to also be a decent terminal based chat client for any production MCP servers too. Requires using your own Anthropic API key, but other than that it should "just work".
Cline is another client. They even have a marketplace of MCP server extensions which you can use with Cline or Claude Desktop. https://cline.bot/mcp-marketplace
One thing I'll say, is that if I was going to make people pay API costs (like cline/claude code) I probably wouldn't actually make an MCP. The MCP box is pretty limiting, and I'm only willing to pay the cost because that's how I get onto flat pricing structure.
I haven't tried it, but another one might be 5ire, seems to be able to use mcp.
You can build your own MCP client https://modelcontextprotocol.io/quickstart/client
You will need an ANTHROPIC_API_KEY though.
Codegen is an MCP client that you can trigger via Slack: codegen.com
It'd be really nice if the AI providers just discounted API tokens against your subscriptions so they were in-effect just pre-paying for (ideally a discounted) set amount of tokens.
Related ongoing thread:
Hacking Your Own AI Coding Assistant with Claude Pro and MCP - https://news.ycombinator.com/item?id=43410866
Nice
btw with claude desktop you can easily hack it to auto-allow MCP, save yourself from clicking "allow" all the time. Dangerous of course if you auto-approve write actions!
https://old.reddit.com/r/ClaudeAI/comments/1h9harx/auto_appr...
Only gripe I have with the tool is that once you've gotten a country right a few times it zooms in too far. I still had no clue where Eritrea was after getting it right like four times. Just got lucky.
But now that the map only shows me three possible countries I can trivially remember which one it was. Ask me again tomorrow while only showing me the full map and I might guess it's in South America.
Hey there, excited first time user here. Thanks for creating this!
I started playing around with it yesterday evening. Running Claude Desktop under manjaro linux with some tweaks, I did managed to activate codemcp.
As I am also quite new with mcp and how to proper use the protocol in tools, I am not sure I did everything correct. I did also install filesystem, git, github and sequential-thinking along side.
However, git will not work. The addition of the mcp git server might not be needed? But without it, it did also not work
However, what it has done so far is mind blowing. I did create a long query the other day about a project idea. This generated a road map, detailed implementation instruction, guidlines, all that is needed to complete the problem. I let it Claude Desktop read in the files with help of codemcp.
Claude Dekstop have been working for multiple hours before I seem to have hot some kind of limit. I now get the message to type continue, but it always start the with the same task.
And what it has been created is awesome. The frontend is nearly finished, but it lacked to create the startup code first. Overall I am blow away what I can do with mcp and esp. with this small lib the author has created here. Code quality is good and event though I had multiple limits, where I needed to type continue, the code is logically make sense. Really good!
I also do not yet know how I can continue a old session.
P.S. @anthropic, If you are reading this. I really do not want to abuse your service. Please, instead of giving me no answer, or beeing stuck in an infinite loop, give me a proper return code and a error, that I reached my limit for the next X minutes or so. I would really love this
I'm surprised you managed to get Desktop running on Linux lol. You don't need the filesystem/git MCPs alongside codemcp; in fact, it's better not to have them so that Claude consistently uses' codemcp's equivalents to do edits. I'm not sure why codemcp's built-in git support did not work; you can probably find out more by looking in .codemcp/codemcp.log.
If you need to start a new chat, it works just fine. Tell Claude what's happened so far and what you want it to do. You can also ask Claude to summarize the old conversation, that's how /compact in claude code works too.
When I posted this to Reddit there was a pretty lively discussion there: https://www.reddit.com/r/ChatGPTCoding/s/wRmnREUWzn
Why are people not using plugins like Cline?!
Cline has famously huge API token costs as it is profligate with context. Because codemcp plugs into Claude Desktop you only pay for your Claude Pro sub, similar to Cursor's pricing model
This is true and is essentially a form of arbitrage. Anthropic is eating the cost of your elevated queries with their $20 flat fee subscription.
The "famously huge API token costs" you are referring to is Cline passing the Anthropic API cost through to you with no markup. You even input your own API token.
Cline also requires the use of an API key, preferably from Anthropic. I found its results to be much better than running a DeepSeek-Llama 70b locally.
I am using Windsurf and consistently running low Flex credits and was looking into MCP for local file editing.
Would be interested in hearing what you learned creating Codemcp, how you find MCP as a protocol, and something else that you found interesting doing your project.
So I have liked the Claude Pro style rate limiting over credits that refresh monthly but mostly because I only get snatches of 1-2 hours while I am on baby leave so I never actually get rate limited. As for learnings, I put a lot of them in my AI Blindspots blog, cuz I did most of codemcp's dev with LLMs
TIL: Model Context Protocol (MCP)
Side question, isn't this against their terms and conditions?
No, in fact, codemcp can be thought of as a fancier version of the official filesystem MCP that Anthropic released. It's 100% MCP.
I've been using it for the past hour or so.
I told it to make a TODO list (based on my instructions) and now I'm just repeatedly creating a new chat just saying 'work on the todo list'. The thing I'm giving it to do is pretty easy (fix some React components to add types and replace hardcoded strings with props). Doing the same thing in Cursor was annoying, due to the chatter, and the fact I had to click 'Accept all'.
For single file changes, it's nice. I haven't tried it with more complex things yet.
In my experience, the model is king, and codemcp will operate fairly similarly to Cursor with the same failure modes as Cursor due to Sonnet 3.7. One thing that I like about codemcp is I can customize aspects of the interaction as I discover new things I want to do :)