tough-tongue/toughtongue-skills

toughtongue-skills is a set of Agent Skills and an MCP server that connect ToughTongue AI to coding assistants like Claude Code, Codex, and Cursor. ToughTongue AI itself is a platform for practicing high stakes conversations, such as interviews, sales calls, coaching sessions, and negotiations, using a voice AI agent that plays the other side of the conversation. This repository is not the ToughTongue platform itself, it is the bridge that lets an AI coding assistant work with your ToughTongue account directly from your terminal or editor. The repository ships two things that work together. Skills are pieces of workflow guidance the assistant loads automatically when it recognizes what you are trying to do. There are four of them: one that verifies your setup after installing and gets you started, one that creates a new practice scenario from a brief, a web link, or a real call transcript, one that fixes an existing scenario by pulling in real session transcripts and finding the smallest change that fixes the problem, and one that turns raw session data into a structured report, such as scores and top improvement areas across a team. The MCP server is the part that actually performs actions in your ToughTongue account: it offers 26 tools for creating and updating scenarios, listing and fetching sessions, pulling analytics, and even placing phone calls or meeting bot sessions. With skills alone the assistant can only give advice, with the MCP server alone it can only run raw tools, installing both through the plugin lets it choose the right workflow and complete it. To use this you need a ToughTongue AI account and, for most setups, a personal access token you create in your account and export as an environment variable, since the README warns never to write that token directly into a config file. Installation differs slightly by tool: Claude Code, Codex, and Cursor each support a one step plugin install that sets up both the skills and the MCP server together, while other agents need the two pieces installed separately. The project has five stars, and the visible portion of the README does not state a license.