0.4.2

• added debouncing to autocompletion in prompt field
• keeping the processing on errors (e.g. LLM model API overload that keeps retrying)
• using --no-cache-dir when installing Python packages on start

0.4.1

• fixed prompt field answer handling to properly prepare for next prompt after answering question
• fixed architect auto-accept behavior in connector to work properly with AiderDesk
• fixed yes/no question answering with custom prompt
• added support for /run command
• added support for /reasoning-effort, /think-tokens commands and showing the values in the project bar
• added Thinking and Answer message blocks when using reasoning models
• fixed watch files infinite loop caused by missing ignores

0.4.0

Highlights

🌐 Control AiderDesk via REST API

We've introduced a REST API to give you more control over context files and project management:

• Run prompts and manage context files directly through the API.
• More flexibility for automation and external tool integrations.

More information

🚀 AiderDesk Gets Its Own MCP Server!

AiderDesk now comes with a built-in MCP server, making it easier than ever to integrate with external tools. You can now:

• Seamlessly manage context files and run prompts via MCP.
• Connect with Claude Desktop, Claude Code, Cursor, Windsurf, and other MCP-supported clients.
• Enjoy a smoother workflow with deep AiderDesk API integration.

More information

Other changes

• added Gemini model support for MCP agent
• enhanced autocompletion in the prompt field, now including abbreviations.
• implemented a fancy loading animation for improved UI feedback.

💡 We’d love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.3.3

• skip adding ignored non read-only files to the context
• improved MCP client interruption handling
• properly adding user input messages to the input history when using MCP tools
• properly wrapping long tool message content
• better handling of MCP tool errors
• increase max buffer size for socket.io events to fix issue with large repos

0.3.2

Hightlights

Configurable system prompt for MCP agent

You can now update the System prompt to better suit the needs of your MCP servers.

Added result of MCP tool to the message

Results of the tool execution are now displayed as part of the Tool message (when expanded).

Added support for ${projectDir} in MCP server configuration

You can use ${projectDir} string in MCP server config inside env and args. This will interpolate the pattern to the current project directory where MCP is executed.

Example:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "${projectDir}"
      ]
    }
  }
}

Other changes

• updated Claude model to 3.7 in default preferred list of models
• fixed prompt field focus issue after model selection
• properly showing preferred models in model selector when searching
• added missing vertical scrollbar when MCP server has many tools

0.3.1

• added /map and /map-refresh commands for repository mapping functionality (#14)
• using python executable to install packages instead of pip (#15)
• prevent infinite loading state after application refresh
• added AIDER_DESK_NO_AUTO_UPDATE environment variable to disable automatic updates

0.3.0

Highlights

Say hi to MCP servers! 🎉

• experimental support for Model Context Protocol (MCP) servers
• connect to external tools like browsers and documentation systems
• choose between OpenAI and Anthropic providers
• enable/disable servers with flexible configuration
• 🚧 Experimental Feature Warning:
  • expect potential instabilities and unexpected behaviors
  • workflow and implementation may change significantly based on user feedback
  • we encourage users to report any issues, suggestions, or use cases
  • future versions will refine the MCP integration based on community input

Other changes

• added support for /test command
• show a loading message in the project bar while models are loading
• fixed duplicated response message after LLM error
• fixed long running message processing (o3-mini, deepseek) to prevent crashes
• updated font family for notifications to match the application
• fixed small model selection issues
• removed an invalid Thinking message

0.2.4

• added proper error message in case there is unrecognized option passed to Aider (#2)
• properly showing the response message for non-streaming models, such as o1

0.2.3

• introduced AIDER_DESK_PYTHON environment variable for an optional Python executable used in the initial setup

0.2.2

• added "Aider frozen" dialog when Aider is unresponsive after interrupt (common case with Deepseek API)
• fixed focus cycling when dialog is open
• fixed autocompletion suggestion popup not rendering properly inside document