v0.13.1

Patch Changes

• #399 734a4fd Thanks @Crunchyman-ralph! - Fix ERR_MODULE_NOT_FOUND when trying to run MCP Server

v0.13.0

Minor Changes

• #240 ef782ff Thanks @eyaltoledano! - feat(expand): Enhance expand and expand-all commands

  • Integrate task-complexity-report.json to automatically determine the number of subtasks and use tailored prompts for expansion based on prior analysis. You no longer need to try copy-pasting the recommended prompt. If it exists, it will use it for you. You can just run task-master update --id=[id of task] --research and it will use that prompt automatically. No extra prompt needed.
  • Change default behavior to append new subtasks to existing ones. Use the --force flag to clear existing subtasks before expanding. This is helpful if you need to add more subtasks to a task but you want to do it by the batch from a given prompt. Use force if you want to start fresh with a task's subtasks.

• #240 87d97bb Thanks @eyaltoledano! - Adds support for the OpenRouter AI provider. Users can now configure models available through OpenRouter (requiring an OPENROUTER_API_KEY) via the task-master models command, granting access to a wide range of additional LLMs. - IMPORTANT FYI ABOUT OPENROUTER: Taskmaster relies on AI SDK, which itself relies on tool use. It looks like free models sometimes do not include tool use. For example, Gemini 2.5 pro (free) failed via OpenRouter (no tool use) but worked fine on the paid version of the model. Custom model support for Open Router is considered experimental and likely will not be further improved for some time.

• #240 1ab836f Thanks @eyaltoledano! - Adds model management and new configuration file .taskmasterconfig which houses the models used for main, research and fallback. Adds models command and setter flags. Adds a --setup flag with an interactive setup. We should be calling this during init. Shows a table of active and available models when models is called without flags. Includes SWE scores and token costs, which are manually entered into the supported_models.json, the new place where models are defined for support. Config-manager.js is the core module responsible for managing the new config."

• #240 c8722b0 Thanks @eyaltoledano! - Adds custom model ID support for Ollama and OpenRouter providers.

  • Adds the --ollama and --openrouter flags to task-master models --set-<role> command to set models for those providers outside of the support models list.
  • Updated task-master models --setup interactive mode with options to explicitly enter custom Ollama or OpenRouter model IDs.
  • Implemented live validation against OpenRouter API (/api/v1/models) when setting a custom OpenRouter model ID (via flag or setup).
  • Refined logic to prioritize explicit provider flags/choices over internal model list lookups in case of ID conflicts.
  • Added warnings when setting custom/unvalidated models.
  • We obviously don't recommend going with a custom, unproven model. If you do and find performance is good, please let us know so we can add it to the list of supported models.

• #240 2517bc1 Thanks @eyaltoledano! - Integrate OpenAI as a new AI provider. - Enhance models command/tool to display API key status. - Implement model-specific maxTokens override based on supported-models.json to save you if you use an incorrect max token value.

• #240 9a48278 Thanks @eyaltoledano! - Tweaks Perplexity AI calls for research mode to max out input tokens and get day-fresh information - Forces temp at 0.1 for highly deterministic output, no variations - Adds a system prompt to further improve the output - Correctly uses the maximum input tokens (8,719, used 8,700) for perplexity - Specificies to use a high degree of research across the web - Specifies to use information that is as fresh as today; this support stuff like capturing brand new announcements like new GPT models and being able to query for those in research. 🔥

Patch Changes

• #240 842eaf7 Thanks @eyaltoledano! - - Add support for Google Gemini models via Vercel AI SDK integration.

• #240 ed79d4f Thanks @eyaltoledano! - Add xAI provider and Grok models support

• #378 ad89253 Thanks @eyaltoledano! - Better support for file paths on Windows, Linux & WSL.

  • Standardizes handling of different path formats (URI encoded, Windows, Linux, WSL).
  • Ensures tools receive a clean, absolute path suitable for the server OS.
  • Simplifies tool implementation by centralizing normalization logic.

• #285 2acba94 Thanks @neno-is-ooo! - Add integration for Roo Code

• #378 d63964a Thanks @eyaltoledano! - Improved update-subtask - Now it has context about the parent task details - It also has context about the subtask before it and the subtask after it (if they exist) - Not passing all subtasks to stay token efficient

• #240 5f504fa Thanks @eyaltoledano! - Improve and adjust init command for robustness and updated dependencies.

  • Update Initialization Dependencies: Ensure newly initialized projects (task-master init) include all required AI SDK dependencies (@ai-sdk/*, ai, provider wrappers) in their package.json for out-of-the-box AI feature compatibility. Remove unnecessary dependencies (e.g., uuid) from the init template.
  • Silence npm install during init: Prevent npm install output from interfering with non-interactive/MCP initialization by suppressing its stdio in silent mode.
  • Improve Conditional Model Setup: Reliably skip interactive models --setup during non-interactive init runs (e.g., init -y or MCP) by checking isSilentMode() instead of passing flags.
  • Refactor init.js: Remove internal isInteractive flag logic.
  • Update init Instructions: Tweak the "Getting Started" text displayed after init.
  • Fix MCP Server Launch: Update .cursor/mcp.json template to use node ./mcp-server/server.js instead of npx task-master-mcp.
  • Update Default Model: Change the default main model in the .taskmasterconfig template.

• #240 96aeeff Thanks @eyaltoledano! - Fixes an issue with add-task which did not use the manually defined properties and still needlessly hit the AI endpoint.

• #240 5aea93d Thanks @eyaltoledano! - Fixes an issue that prevented remove-subtask with comma separated tasks/subtasks from being deleted (only the first ID was being deleted). Closes #140

• #240 66ac9ab Thanks @eyaltoledano! - Improves next command to be subtask-aware - The logic for determining the "next task" (findNextTask function, used by task-master next and the next_task MCP tool) has been significantly improved. Previously, it only considered top-level tasks, making its recommendation less useful when a parent task containing subtasks was already marked 'in-progress'. - The updated logic now prioritizes finding the next available subtask within any 'in-progress' parent task, considering subtask dependencies and priority. - If no suitable subtask is found with...

v0.12.1

task-master-ai

0.12.0

Minor Changes

• #253 b2ccd60 Thanks @Crunchyman-ralph! - Add npx task-master-ai that runs mcp instead of using `task-master-mcp``

• #267 c17d912 Thanks @Crunchyman-ralph! - Improve PRD parsing prompt with structured analysis and clearer task generation guidelines. We are testing a new prompt - please provide feedback on your experience.

Patch Changes

• #243 454a1d9 Thanks @Crunchyman-ralph! - - Fixes shebang issue not allowing task-master to run on certain windows operating systems

  • Resolves #241 #211 #184 #193

• #268 3e872f8 Thanks @Crunchyman-ralph! - Fix remove-task command to handle multiple comma-separated task IDs

• #239 6599cb0 Thanks @eyaltoledano! - Updates the parameter descriptions for update, update-task and update-subtask to ensure the MCP server correctly reaches for the right update command based on what is being updated -- all tasks, one task, or a subtask.

• #272 3aee9bc Thanks @Crunchyman-ralph! - Enhance the parsePRD to include --append flag. This flag allows users to append the parsed PRD to an existing file, making it easier to manage multiple PRD files without overwriting existing content.

• #264 ff8e75c Thanks @joedanz! - Add quotes around numeric env vars in mcp.json (Windsurf, etc.)

• #248 d99fa00 Thanks @Crunchyman-ralph! - - Fix task-master init polluting codebase with new packages inside package.json and modifying project README

  • Now only initializes with cursor rules, windsurf rules, mcp.json, scripts/example_prd.txt, .gitignore modifications, and README-task-master.md

• #266 41b979c Thanks @Crunchyman-ralph! - Fixed a bug that prevented the task-master from running in a Linux container

• #265 0eb16d5 Thanks @Crunchyman-ralph! - Remove the need for project name, description, and version. Since we no longer create a package.json for you

0.11.0

Minor Changes

• #71 7141062 Thanks @eyaltoledano! - - Easier Ways to Use Taskmaster (CLI & MCP):
  • You can now use Taskmaster either by installing it as a standard command-line tool (task-master) or as an MCP server directly within integrated development tools like Cursor (using its built-in features). This makes Taskmaster accessible regardless of your preferred workflow.
  • Setting up a new project is simpler in integrated tools, thanks to the new initialize_project capability.
  • Complete MCP Implementation:
    • NOTE: Many MCP clients charge on a per tool basis. In that regard, the most cost-efficient way to use Taskmaster is through the CLI directly. Otherwise, the MCP offers the smoothest and most recommended user experience.
    • All MCP tools now follow a standardized output format that mimicks RESTful API responses. They are lean JSON responses that are context-efficient. This is a net improvement over the last version which sent the whole CLI output directly, which needlessly wasted tokens.
    • Added a remove-task command to permanently delete tasks you no longer need.
    • Many new MCP tools are available for managing tasks (updating details, adding/removing subtasks, generating task files, setting status, finding the next task, breaking down complex tasks, handling dependencies, analyzing complexity, etc.), usable both from the command line and integrated tools. (See the taskmaster.mdc reference guide and improved readme for a full list).
  • Better Task Tracking:
    • Added a "cancelled" status option for tasks, providing more ways to categorize work.
  • Smoother Experience in Integrated Tools:
    • Long-running operations (like breaking down tasks or analysis) now run in the background via an Async Operation Manager with progress updates, so you know what's happening without waiting and can check status later.
  • Improved Documentation:
    • Added a comprehensive reference guide (taskmaster.mdc) detailing all commands and tools with examples, usage tips, and troubleshooting info. This is mostly for use by the AI but can be useful for human users as well.
    • Updated the main README with clearer instructions and added a new tutorial/examples guide.
    • Added documentation listing supported integrated tools (like Cursor).
  • Increased Stability & Reliability:
    • Using Taskmaster within integrated tools (like Cursor) is now more stable and the recommended approach.
    • Added automated testing (CI) to catch issues earlier, leading to a more reliable tool.
    • Fixed release process issues to ensure users get the correct package versions when installing or updating via npm.
  • Better Command-Line Experience:
    • Fixed bugs in the expand-all command that could cause NaN errors or JSON formatting issues (especially when using --research).
    • Fixed issues with parameter validation in the analyze-complexity command (specifically related to the threshold parameter).
    • Made the add-task command more consistent by adding standard flags like --title, --description for manual task creation so you don't have to use --prompt and can quickly drop new ideas and stay in your flow.
    • Improved error messages for incorrect commands or flags, making them easier to understand.
    • Added confirmation warnings before permanently deleting tasks (remove-task) to prevent mistakes. There's a known bug for deleting multiple tasks with comma-separated values. It'll be fixed next release.
    • Renamed some background tool names used by integrated tools (e.g., list-tasks is now get_tasks) to be more intuitive if seen in logs or AI interactions.
    • Smoother project start: Improved the guidance provided to AI assistants immediately after setup (related to init and parse-prd steps). This ensures the AI doesn't go on a tangent deciding its own workflow, and follows the exact process outlined in the Taskmaster workflow.
  • Clearer Error Messages:
    • When generating subtasks fails, error messages are now clearer, including specific task IDs and potential suggestions.
    • AI fallback from Claude to Perplexity now also works the other way around. If Perplexity is down, will switch to Claude.
  • Simplified Setup & Configuration:
    • Made it clearer how to configure API keys depending on whether you're using the command-line tool (.env file) or an integrated tool (.cursor/mcp.json file).
    • Taskmaster is now better at automatically finding your project files, especially in integrated tools, reducing the need for manual path settings.
    • Fixed an issue that could prevent Taskmaster from working correctly immediately after initialization in integrated tools (related to how the MCP server was invoked). This should solve the issue most users were experiencing with the last release (0.10.x)
    • Updated setup templates with clearer examples for API keys.
    • **For advanced users setting up the MCP server manually, the command is now npx -y task-master-ai task-master-mcp.
  • Enhanced Performance & AI:
    • Updated underlying AI model settings:
      • Increased Context Window: Can now handle larger projects/tasks due to an increased Claude context window (64k -> 128k tokens).
      • Reduced AI randomness: More consistent and predictable AI outputs (temperature 0.4 -> 0.2).
      • Updated default AI models: Uses newer models like claude-3-7-sonnet-20250219 and Perplexity sonar-pro by default.
      • More granular breakdown: Increas...

v0.11.1

task-master-ai

0.11.0

Minor Changes

• #71 7141062 Thanks @eyaltoledano! - - Easier Ways to Use Taskmaster (CLI & MCP):
  • You can now use Taskmaster either by installing it as a standard command-line tool (task-master) or as an MCP server directly within integrated development tools like Cursor (using its built-in features). This makes Taskmaster accessible regardless of your preferred workflow.
  • Setting up a new project is simpler in integrated tools, thanks to the new initialize_project capability.
  • Complete MCP Implementation:
    • NOTE: Many MCP clients charge on a per tool basis. In that regard, the most cost-efficient way to use Taskmaster is through the CLI directly. Otherwise, the MCP offers the smoothest and most recommended user experience.
    • All MCP tools now follow a standardized output format that mimicks RESTful API responses. They are lean JSON responses that are context-efficient. This is a net improvement over the last version which sent the whole CLI output directly, which needlessly wasted tokens.
    • Added a remove-task command to permanently delete tasks you no longer need.
    • Many new MCP tools are available for managing tasks (updating details, adding/removing subtasks, generating task files, setting status, finding the next task, breaking down complex tasks, handling dependencies, analyzing complexity, etc.), usable both from the command line and integrated tools. (See the taskmaster.mdc reference guide and improved readme for a full list).
  • Better Task Tracking:
    • Added a "cancelled" status option for tasks, providing more ways to categorize work.
  • Smoother Experience in Integrated Tools:
    • Long-running operations (like breaking down tasks or analysis) now run in the background via an Async Operation Manager with progress updates, so you know what's happening without waiting and can check status later.
  • Improved Documentation:
    • Added a comprehensive reference guide (taskmaster.mdc) detailing all commands and tools with examples, usage tips, and troubleshooting info. This is mostly for use by the AI but can be useful for human users as well.
    • Updated the main README with clearer instructions and added a new tutorial/examples guide.
    • Added documentation listing supported integrated tools (like Cursor).
  • Increased Stability & Reliability:
    • Using Taskmaster within integrated tools (like Cursor) is now more stable and the recommended approach.
    • Added automated testing (CI) to catch issues earlier, leading to a more reliable tool.
    • Fixed release process issues to ensure users get the correct package versions when installing or updating via npm.
  • Better Command-Line Experience:
    • Fixed bugs in the expand-all command that could cause NaN errors or JSON formatting issues (especially when using --research).
    • Fixed issues with parameter validation in the analyze-complexity command (specifically related to the threshold parameter).
    • Made the add-task command more consistent by adding standard flags like --title, --description for manual task creation so you don't have to use --prompt and can quickly drop new ideas and stay in your flow.
    • Improved error messages for incorrect commands or flags, making them easier to understand.
    • Added confirmation warnings before permanently deleting tasks (remove-task) to prevent mistakes. There's a known bug for deleting multiple tasks with comma-separated values. It'll be fixed next release.
    • Renamed some background tool names used by integrated tools (e.g., list-tasks is now get_tasks) to be more intuitive if seen in logs or AI interactions.
    • Smoother project start: Improved the guidance provided to AI assistants immediately after setup (related to init and parse-prd steps). This ensures the AI doesn't go on a tangent deciding its own workflow, and follows the exact process outlined in the Taskmaster workflow.
  • Clearer Error Messages:
    • When generating subtasks fails, error messages are now clearer, including specific task IDs and potential suggestions.
    • AI fallback from Claude to Perplexity now also works the other way around. If Perplexity is down, will switch to Claude.
  • Simplified Setup & Configuration:
    • Made it clearer how to configure API keys depending on whether you're using the command-line tool (.env file) or an integrated tool (.cursor/mcp.json file).
    • Taskmaster is now better at automatically finding your project files, especially in integrated tools, reducing the need for manual path settings.
    • Fixed an issue that could prevent Taskmaster from working correctly immediately after initialization in integrated tools (related to how the MCP server was invoked). This should solve the issue most users were experiencing with the last release (0.10.x)
    • Updated setup templates with clearer examples for API keys.
    • **For advanced users setting up the MCP server manually, the command is now npx -y task-master-ai task-master-mcp.
  • Enhanced Performance & AI:
    • Updated underlying AI model settings:
      • Increased Context Window: Can now handle larger projects/tasks due to an increased Claude context window (64k -> 128k tokens).
      • Reduced AI randomness: More consistent and predictable AI outputs (temperature 0.4 -> 0.2).
      • Updated default AI models: Uses newer models like claude-3-7-sonnet-20250219 and Perplexity sonar-pro by default.
      • More granular breakdown: Increased the default number of subtasks generated by expand to 5 (from 4).
      • Consistent defaults: Set the default priority for new tasks consistently to "medium".
    • Improved performance when viewing task details in integrated tools by sending less redundant data.
  • Documentation Clarity:
    • Clarified in documentation that Markdown files (.md) can be used for Product Requirements Documents (parse_prd).
    • Improved the description for the numTasks option in parse_prd for better guidance.
  • Improved Visuals (CLI):
    • Enhanced the look and feel of progress bars and status updates in the command line.
    • Added a helpful color-coded progress bar to the task details view (show command) to visualize subtask completion.
    • Made progress bars show a breakdown of task statuses (e.g., how many are pending vs. done).
    • Made status counts clearer with text labels next to icons.
    • Prevented progress bars from messing up the display on smaller terminal windows.
    • Adjusted how progress is calculated for 'deferred' and 'cancelled' tasks in the progress bar, while still showing their distinct status visually.
  • Fixes for Integrated Tools:
    • Fixed how progress updates are sent to integrated tools, ensuring they display correctly.
    • Fixed internal issues that could cause errors or invalid JSON responses when using Taskmaster with integrated tools.

0.10.1

Patch Changes

• #80 aa185b2 Thanks @Crunchyman-ralph! - Remove non-existent package @model-context-protocol/sdk

• #45 757fd47 Thanks @Crunchyman-ralph! - Add license to repo

0.10.0

Minor Changes

• #44 eafdb47 Thanks @Crunchyman-ralph! - add github actions to automate github and npm releases

• #20 4eed269 Thanks @Crunchyman-ralph! - Implement MCP server for all commands using tools.

Patch Changes

• #44 44db895 Thanks @Crunchyman-ralph! - Added changeset config #39

• #50 257160a Thanks @Crunchyman-ralph! - Fix addTask tool projectRoot not defined

• #57 9fd42ee Thanks @github-actions! - fix mcp server not connecting to cursor

• #48 5ec3651 Thanks @Crunchyman-ralph! - Fix workflows

v0.10.2

task-master-ai

0.10.1

Patch Changes

• #80 aa185b2 Thanks @Crunchyman-ralph! - Remove non-existent package @model-context-protocol/sdk

• #45 757fd47 Thanks @Crunchyman-ralph! - Add license to repo

0.10.0

Minor Changes

• #44 eafdb47 Thanks @Crunchyman-ralph! - add github actions to automate github and npm releases

• #20 4eed269 Thanks @Crunchyman-ralph! - Implement MCP server for all commands using tools.

Patch Changes

• #44 44db895 Thanks @Crunchyman-ralph! - Added changeset config #39

• #50 257160a Thanks @Crunchyman-ralph! - Fix addTask tool projectRoot not defined

• #57 9fd42ee Thanks @github-actions! - fix mcp server not connecting to cursor

• #48 5ec3651 Thanks @Crunchyman-ralph! - Fix workflows

v0.9.30

https://www.npmjs.com/package/task-master-ai/v/0.9.30

This is mostly a chore release solving a few bugs and is heavily focused on implementing unit tests for the Task Master platform. We are off to a good start there with nearly 200 tests defined and 169 implemented and passing.

This release paves the way for the next which will be a minor patch introducing MCP support. Will be holding minor improvements, bug fixes and everything else until that ships as it is the number requested feature at the moment and the most likely feature to help reduce friction getting started with Task Master.

How to upgrade
Get the latest features by running npm i -g task-master-ai and then run task-master init in your project root folder. Tasks will not be affected by the init command as it updates other files.

New features:

• feature: Enabled 128k token output for Claude 3.7 Sonnet commit
• feature: Adds add-subtasks and remove-subtasks commands for manually adding subtasks. Can also turn subtasks into standalone features commit
• feature: Improve parse-prd command commit
  • Adds default PRD path support (scripts/prd.txt) so you can just run task-master parse-prd and it will use the default PRD if it exists
• feature: Adds .windsurfrules to the init package. commit
  • The Windsurf rules are composed of the 3 rules we currently package, and has been edited to be Windsurf specific. Rules are added in as sections. The init function will search for an existing .windsurfrules document, and if it finds it, it will append to it. Otherwise it will create it.

Bug fixes:

• fix: Tweak table column widths. Will probably make them dynamicalyl adjust based on the longest string in the column. But that's an over-optimization for now. commit
• fix: implements better config flag support to task-master init commit
• fix: elegantly exit if running into a claude error like overloaded api + integration test. commit
• fix: subtask id is truncated in task show subtask table. commit
• fix: ensure CLI correctly handles kebab-case options commit
• fix: Claude Streaming bug commit

New Contributors

• @eyaltoledano created the project
• @Crunchyman-ralph made their first contribution in #1