CodeMesh helps AI coding tools like Claude Code, Cursor, and Windsurf understand your code faster. Instead of reading the same files again and again, CodeMesh gives them only the code they need. This can reduce token usage and AI costs by over 90%, while making coding agents faster and more efficient. You keep using your favorite coding tool CodeMesh simply makes it smarter and cheaper.
Hey Product Hunt :wave: I’m Vamsi, from the team of CodeMesh.
CodeMesh started because my AI coding bill got much higher than I expected. I was using Claude Code and Cursor every day, and while they were great at writing code, I noticed that a lot of tokens were not being spent on actually solving the problem. They were being spent searching the repo, opening large files, reading thousands of lines, following imports, finding where functions were used, and then reading the same code again on the next task. Basically, I was paying expensive AI models to search through my codebase over and over again.
So we built CodeMesh. CodeMesh understands the structure of your codebase once and creates a live map of it. When Claude Code, Cursor, Windsurf, or another coding agent needs to understand your code, CodeMesh gives it only the relevant information instead of making the agent read huge parts of the repository again. Instead of saying, “Read these 3,000 lines and figure it out,” CodeMesh can tell the agent where a function is, what calls it, what it imports, and which files are related. This means your coding agent spends fewer tokens looking for code and more tokens actually solving your task.
We tested CodeMesh using 98 real software engineering questions on the public OpenClaw repository. We used the same codebase, the same questions, and the same model for both runs. Without CodeMesh, the agent used 25,295,273 tokens. With CodeMesh, it used 1,227,371 tokens. That’s 95.1% fewer tokens. Measured API cost dropped from $23.67 to $1.65, which is 93.1% lower. You can see the full benchmark here: https://codemesh.cloud/benchmarks
Under the hood, CodeMesh creates a live graph of your code. It understands functions, classes, imports, function calls, dependencies, and relationships between files. So instead of repeatedly reading your entire repository, your AI agent can ask CodeMesh questions about the code and get a precise answer. CodeMesh actually parses the structure of your code rather than only relying on embeddings or similarity search.
The best part is that you don’t need to change the coding tools you already use. CodeMesh works underneath them through MCP. You can keep using Claude Code, Cursor, Windsurf, VS Code, Claude.ai, ChatGPT, or any MCP-compatible agent.
CodeMesh simply gives those tools better context so they can work with fewer tokens.
We also designed CodeMesh to be read-only. Your agents can use it to understand your repository, but they cannot modify your code through the CodeMesh context graph. Authentication and organization isolation are built in because security was one of the first things engineering teams asked us about.
Getting started is simple. Signup, install extension, connect your coding agent, and continue working the way you already do.
For our Product Hunt launch, for the next 7 days, connect a repository and we’ll run the same benchmark on your own codebase for free. We’ll show you how many tokens and how much AI cost you could save using CodeMesh
If you use Claude Code, Cursor, Windsurf, or other AI coding agents regularly, we’d love for you to try CodeMesh and tell us what you think -> https://codemesh.cloud/signup
what happens with code that gets called by a string name, like event handlers?
Reducing coding-agent token usage through structural code graphs is a promising approach. How does CodeMesh handle cross-repository dependencies in microservice architectures, and can agents trace function calls across multiple repositories?
Congrats on CodeMesh. Mine is a handful of repos I open once a month each, so every session begins with the agent reading its way back in. Is the graph meant to cover that, a repo it has not seen since August?
The OpenClaw benchmark is what sold me — 25M tokens down to 1.2M on the same 98 questions is a much more honest test than a cherry-picked demo. I'm curious how the live graph holds up as a session goes on, though: when the agent edits files mid-task, does the map update right away, or can it end up pointing the agent at code that's already changed?
This product literally caught me at a time where I am maxing 3 Claude Subscriptions daily. Fantastic idea.
About CodeMesh on Product Hunt
“Cut AI coding costs by 90%”
CodeMesh launched on Product Hunt on November 14th, 2026 and earned 81 upvotes and 10 comments, placing #17 on the daily leaderboard. CodeMesh helps AI coding tools like Claude Code, Cursor, and Windsurf understand your code faster. Instead of reading the same files again and again, CodeMesh gives them only the code they need. This can reduce token usage and AI costs by over 90%, while making coding agents faster and more efficient. You keep using your favorite coding tool CodeMesh simply makes it smarter and cheaper.
CodeMesh was featured in Software Engineering (43.3k followers), Developer Tools (523.3k followers) and Vibe coding (774 followers) on Product Hunt. Together, these topics include over 104.3k products, making this a competitive space to launch in.
Who hunted CodeMesh?
CodeMesh was hunted by Vamsi Krishna. A “hunter” on Product Hunt is the community member who submits a product to the platform — uploading the images, the link, and tagging the makers behind it. Hunters typically write the first comment explaining why a product is worth attention, and their followers are notified the moment they post. Around 79% of featured launches on Product Hunt are self-hunted by their makers, but a well-known hunter still acts as a signal of quality to the rest of the community. See the full all-time top hunters leaderboard to discover who is shaping the Product Hunt ecosystem.
Want to see how CodeMesh stacked up against nearby launches in real time? Check out the live launch dashboard for upvote speed charts, proximity comparisons, and more analytics.
Hey Product Hunt :wave: I’m Vamsi, from the team of CodeMesh.
CodeMesh started because my AI coding bill got much higher than I expected. I was using Claude Code and Cursor every day, and while they were great at writing code, I noticed that a lot of tokens were not being spent on actually solving the problem. They were being spent searching the repo, opening large files, reading thousands of lines, following imports, finding where functions were used, and then reading the same code again on the next task. Basically, I was paying expensive AI models to search through my codebase over and over again.
So we built CodeMesh. CodeMesh understands the structure of your codebase once and creates a live map of it. When Claude Code, Cursor, Windsurf, or another coding agent needs to understand your code, CodeMesh gives it only the relevant information instead of making the agent read huge parts of the repository again. Instead of saying, “Read these 3,000 lines and figure it out,” CodeMesh can tell the agent where a function is, what calls it, what it imports, and which files are related. This means your coding agent spends fewer tokens looking for code and more tokens actually solving your task.
We tested CodeMesh using 98 real software engineering questions on the public OpenClaw repository. We used the same codebase, the same questions, and the same model for both runs. Without CodeMesh, the agent used 25,295,273 tokens. With CodeMesh, it used 1,227,371 tokens. That’s 95.1% fewer tokens. Measured API cost dropped from $23.67 to $1.65, which is 93.1% lower. You can see the full benchmark here: https://codemesh.cloud/benchmarks
Under the hood, CodeMesh creates a live graph of your code. It understands functions, classes, imports, function calls, dependencies, and relationships between files. So instead of repeatedly reading your entire repository, your AI agent can ask CodeMesh questions about the code and get a precise answer. CodeMesh actually parses the structure of your code rather than only relying on embeddings or similarity search.
The best part is that you don’t need to change the coding tools you already use. CodeMesh works underneath them through MCP. You can keep using Claude Code, Cursor, Windsurf, VS Code, Claude.ai, ChatGPT, or any MCP-compatible agent.
CodeMesh simply gives those tools better context so they can work with fewer tokens.
We also designed CodeMesh to be read-only. Your agents can use it to understand your repository, but they cannot modify your code through the CodeMesh context graph. Authentication and organization isolation are built in because security was one of the first things engineering teams asked us about.
Getting started is simple. Signup, install extension, connect your coding agent, and continue working the way you already do.
For our Product Hunt launch, for the next 7 days, connect a repository and we’ll run the same benchmark on your own codebase for free. We’ll show you how many tokens and how much AI cost you could save using CodeMesh
If you use Claude Code, Cursor, Windsurf, or other AI coding agents regularly, we’d love for you to try CodeMesh and tell us what you think -> https://codemesh.cloud/signup