Ok, ok, before anyone says anything: yes, AI can be bad, but it can also be good. This post was not written by AI (it did review it, mostly for spelling and grammar). Some people think I don’t have any intelligence either… but that’s a different story.
Anyway, I’ve been using Claude Code and Desktop for a while to help with different personal projects, and I’ve added a few things to my setup to help with tasks. Here’s the list of what I use and what each one does.
- jCodeMunch: Indexes your repo so the agent can pull just the function or class it needs instead of reading whole files. It helps organise large repos and makes it easier to find things in a project. Handy bit of kit. They claim it “cuts code-reading tokens by as much as 95% or more.” Free for non-commercial use; commercial use needs a one-time licence (from $79 for one developer). Haven’t purchased a licence yet, but plan to get it soon.
- Context7: An MCP server that gives your AI agent up-to-date documentation for the libraries it’s working with. The free plan covers public libraries and 1,000 API calls a month; Pro ($10 per seat per month) adds private repos and gives 2,000 API calls. Pricing page has more details.
- Headroom: Sits between your AI agent and the model endpoint and compresses tool outputs, logs and files locally before they’re sent. It’s available as a library, proxy or MCP server, and it’s open source (Apache 2.0). I’ve been running it for the last week or so and it has saved me about 5.6 million tokens so far. Heads up: it sends anonymous usage stats by default. Set
HEADROOM_BEACON=offto disable them. - rtk: A single Rust binary that hooks into your agent and rewrites shell commands (for example,
git statusbecomesrtk git status) so the model only sees compact output, reducing the data sent to the AI endpoint. Also open source (Apache 2.0). Note that it only filters Bash commands, not Claude Code’s built-in Read/Grep tools, and the “up to 90%” figure is a cut in command output, not your whole bill.
That’s what I’m currently using, but I’ll probably add more over time. If you’re using something I’ve missed, let me know.
Pricing and licence details as of September 2026.
