just shipped taskmaster v0.12 🚀 → parse multiple PRDs, smarter breakdowns, easier setup, command improvements,MCP polish +++ → introducing the npm i task-master-ai website, which i have become ossessed with (link below) → 4.5k ⭐ GitHub big upgrades based on your feedback 👇

1 you can now parse multiple PRDs without losing previous tasks, introducing batch adding new tasks and prd-by-prd task management just use parse-prd with the --append flag ✅ add tasks by milestone, by feature, by version do it however you structure your product approach

2 next, we added chain-of-thought to PRD parsing 🧠 tasks generated from your PRDs should now be even clearer, better structured and closer to your PRD details still recommend upping task # if prd is complex this is an experimental prompt so share feedback if you notice smth

3 improves MCP onboarding and solve an issue that required some folks to install the npm package globally for mcp to work now it'll work with just: npx -y task-master-ai in your MCP config (don't forget your API keys). clean. btw you'll still need .env if using CLI commands

4 cleaner init: `task-master init` no longer pollutes your codebase with unnecessary files and dependencies in your package.json. we don't touch em anymore. now init ships with: - @cursor_ai + @windsurf_ai rules - mcp.json - scripts/example_prd.tx - gitignore adds - readme

5 small but mighty fixes: - multi-task removals now supported (remove-task) with comma-separated --id=1,2,3 - improved MCP calls to update, update-task, & update-subtask - better env support for clients like @windsurf_ai @AnthropicAI - shebang fix for linux/WSL environments 🛠️

your feedback shaped this update, and we want more. try v0.12 and let us know how it's working for you. next: we're excited to introduce the home of taskmaster's open source CLI and MCP package and your source of info and product roadmap 👇👇

it's a fully interactive CLI in which you can type in task-master commands to understand how they work there's a bunch of commands for you to explore and some frustration prompts to make you feel seen 🫂 + a few easter eggs to discover in the spirit of easter + more coming

that's it for now! if you're enjoying or getting value from using @TaskmasterAI, consider contributing or support the project on github, sharing it with your friends and colleagues, and sending your raw, unfiltered feedback so we can keep improving until then, vibe on friends

full changelog on @github

help taskmaster reach more people by retweeting the first tweet

@TaskmasterAI @github all contributions are welcome, we're swimming in requests rn

@TaskmasterAI @github Do you have tools you are using to create AI friendly PRDs?

@TaskmasterAI @github mostly freestyling, but this is coming

@TaskmasterAI @github Goat 🐐

@TaskmasterAI @github Website is awesome

@TaskmasterAI @github 🙏

@TaskmasterAI @github Just looked at the MCP install. I have Perplexity Pro and was a big fan of it, but my last prompts to Gemini Pro 2.5 returned so much better results, can I just swap them? Also, do I have to have direct API access to the model provider or can @OpenRouterAI be used as well ?

@TaskmasterAI @github @OpenRouterAI it’s coming ser

@TaskmasterAI @github @OpenRouterAI 😍👍🙏

@TaskmasterAI @github Congrats on the big update! Taskmaster just keeps getting better. The website obsession is totally justified—great work!

@TaskmasterAI @github thanks so much terry!

@TaskmasterAI @github 👏👏👏

@TaskmasterAI @github 🙌🏻

@TaskmasterAI @github taskmaster is amazing thanks for building this bro

@TaskmasterAI @github thanks for using it. send feedback!

@TaskmasterAI @github Is there a way to just use the models being used by the IDE (Cursor, Windsurf, etc) to do the tasks where Claude or Perplexity are needed? Since we are already paying for that, why not also let the Developers decide if they want to leverage the IDE agentic capabilities of BYOK?

@TaskmasterAI @github you're referring to 'sampling' which is an MCP feature that's planned yep

@TaskmasterAI How do I configure TaskmasterAI to use Github Copilot in Visual Studio Code instead of Anthropic Api Keys?

@TaskmasterAI you currently can’t - but this is a planned improvement

@TaskmasterAI @github I might have missed it but what is your recommended workflow for PRD including it's granularity?

@TaskmasterAI @github what do you mean?

@TaskmasterAI @github the website is 🔥

@TaskmasterAI @github 🙏

@TaskmasterAI @github Based tool. Can’t wait to try this out

@TaskmasterAI @github how do we migrate from previous versions? Do I have to reinstall?

@TaskmasterAI @github You can just npm i task-master-ai@latest Then do task-master init to bring the new files into your project’s node modules

@TaskmasterAI @github got it. btw do you guys ever plan to support other package managers like bun or pnpm? obviously i can install using them, but the task master still uses npm i insternally, so it creates package lock which is undesirable

@TaskmasterAI @github i would LOVE to support bun AND pnpm i just have no idea how and haven't really prioritized it

@TaskmasterAI @github I’ll look into it and will try to contribute for sure! The bun/pnpm support is crucial for me. I like task-master so much that right now my working directory is one layer above my actual codebase and its just for the task-master, would love to move down one layer

@TaskmasterAI @github with you there - that would be amazing!

@TaskmasterAI @github How would I use this with @lovable_dev?

@TaskmasterAI @github @lovable_dev Through the terminal or the agent - long as it has access to MCP or able to execute terminal commands you’re good

@TaskmasterAI @github Congratz on the release! Let me try it out to current working project

@TaskmasterAI @github Thanks for merging my prompt

Another great update @EyalToledano Would love to see @TaskmasterAI log task completions, issues, discoveries in some kind of implantation journal to provide context for learning and new rules. Immediate and always on project context would really help prevent missteps and redundant tool calls

@TaskmasterAI @github well received - something like a taskmaster.log?

@TaskmasterAI Great, thanks, still on top of my "use for the next project". You still recommend Cursor + Sonnet or can it work with Gemini 2.5 Pro/Flash (equally) as well ?

@TaskmasterAI once we release support for other models I’ll know more but right now Cursor/TM/3.7/Perplexity is an insane stack

@TaskmasterAI @github moving fast mate, will try your mcp today

@TaskmasterAI @github Keep cooking man. This is awesome.

@TaskmasterAI @github

@TaskmasterAI @github Congratulations! I look forward to playing with this today

@TaskmasterAI @github props goes to the rest of the team who pushed this through especially @RalphEcom - i been underwater upgrading taskmaster in my bubble lol

@TaskmasterAI @github Is there any reason this would not work with Claude code?

@TaskmasterAI @github nope, in fact it works great with claude code either using the CLI commands or with the MCP server

@TaskmasterAI @github Nice!!! Thank you for creating this

@TaskmasterAI @github pleasure -- send feedback!

@TaskmasterAI @github will do! I started using it for @orakemu with windsurf so it's already quite a big project - i'll let you know if i have any ideas or feedback

@TaskmasterAI @github Is there anything important to keep in mind when updating existing projects? What is the best way to make use of the new task generation to improve existing tasks?

@TaskmasterAI @github @grok what is it?

@TaskmasterAI @github Awesome! How do I configure it to work with a remote session?

@TaskmasterAI @github It depends on the MCP client based on my experience session.roots has different values depending on the MCP client 😂 should be looking for the files in the project root, but that too will depend on how the config of the mcp client works. roots are annoying

Thanks. I'm trying to use it in Cursor. Do you know how I can set the session.roots? I just get 'It seems the initialization failed with an ENOENT: no such file or directory error when trying to change directory. This might indicate an issue with the path provided or permissions.'

@TaskmasterAI @github hit me in dm’s or on discord, better for support you SHOULD be able to pass a projectRoot argument to init and parse prd but not 100% sure What does the mcp json look like?

@TaskmasterAI @github Ok cool. So change the args to ["-y", "task-master-ai", "projectRoot", "user@host:/path"] ?

@TaskmasterAI @github No leave the mcp.json as is When running the initialize mcp tool, ask the agent to throw in the current projectRoot as an arg for the tool itself the value of that arg is prioritized over the session roots stuff, which should work provided the LLM throws in the current root

@TaskmasterAI @github Ok cool. It seems that it already had that context but it's not able to access it. I'm guessing because the mcp server is running on my machine and not the remote one, so it tries to use the path locally which of course doesn't exist.

@TaskmasterAI @github Yep that makes sense session.roots man, you gotta love em cc @RalphEcom

@TaskmasterAI @github what are you using for editing mate?

@TaskmasterAI @github screen studio

@TaskmasterAI @github ❤️

@TaskmasterAI @github Great tool. though Claude is PROHIBITIVE for large workflows. Why would your initial direction not be a multi model solution? ; you risk a vampire-like improvement fork attack

@TaskmasterAI @github it’s wip

Yes I understand that and i have seen your progress here : . But at least put out a semver patch with openrouter or requesty working API and then improve from there. I have sat on your tool for the last 3 weeks due to this your limitation. Great work and attention to detail by the way. I like what you did there with the terminal visual architecture and aesthetics. Kudos!

i get where you’re coming from but a “patch” for one provider is a patch for all providers. i’m not going to monkey-patch providers one on top of the other and make the env a living hell to deal with i’m adjusting so we can do them properly otherwise i’ll have to come back to this again, which means slowing down other initiatives in the future doesn’t have to be perfect, but it has to work well enough so I don’t have to come back to it every release it’s 80% there. wont be long

Understood and it makes sense; for modularity and future-proofing sake its cool, but your audience awaits to launch you to the stratosphere... and the vultures wait also... let them not catch a wif. But then again imitation is validity. Your tool is a very strong candidate for inclusion in my long list of tools. So if I may ask how long was your 80% so i might be able to estimate the 20% time left ?

@TaskmasterAI @github why on earth it is still anthropic only?

@TaskmasterAI @github my fingers they can only type so fast blame the fingers

@TaskmasterAI @github come ooon dude you have the taskmaster :)) kudos for all the great work btw, don't think I'm ungrateful. just prefer gemini :)

@TaskmasterAI @github it's wip :D

Here I share an XGBoost model that delivers a 25% CAGR with minimal drawdown on Visa stock. In this free Substack post I share code and commentary for a powerful Machine Learning strategy that delivers powerful returns.