Jan 8
Tool-calling via MCP
Enable Superwhisper to connect to MCP servers via configuration. This allows the LLM to move beyond transcription by enabling tool-calling to execute actions across local and cloud services (e.g., Jira, GitHub, Slack, or local file systems).
Voice-to-Action: Transforms Superwhisper from a passive transcription tool into an active agent capable of executing commands (e.g., "Create a Trello card for this").
Instant Ecosystem: Supporting the MCP standard provides immediate access to a growing library of community-built integrations without requiring bespoke development for each service.
Local-First Extensibility: MCP supports local server transports (stdio), aligning with Superwhisper’s privacy-centric, local-first architecture.
Contextual Awareness: Allows the LLM to pull in external resources (codebases, docs) to improve transcription accuracy and provide more relevant responses.
Proposed Implementation
Configuration: Add an "MCP Servers" section in settings to define server paths or endpoints.
Workflow: Fetch tool schemas from configured servers and include them in the LLM prompt, enabling seamless tool-calling during dictation. Enabling/disabling tools would be important for minimizing context window usage.
Pending
