Tough Tongue AI Skills

Agent skills and MCP server for Tough Tongue AI, the platform for handling tough conversations. Some, the AI takes: voice agents that answer and place calls, qualify leads, run demos, screen candidates, and book meetings. Others, you nail: hyper-realistic roleplay that gets you ready for negotiations, interviews, and coaching conversations.

Install once, then work with Tough Tongue AI from Claude Code, Codex, Cursor, or any agent that supports the Agent Skills format:

• "Pull the last 3 lost deals from our call notes and create a practice scenario for the pricing objection."
• "For the Enterprise Discovery Call scenario, pull the last 50 sessions. What are the top 5 improvement areas?"
• "The onboarding agent keeps ending calls too early. Pull the low-scoring sessions, find out why, and fix the scenario."

These compose with the rest of your MCP ecosystem: Gong, Notion, calendar, slides, email. See What you can do for the full journeys.

Table of contents

• What's included
• How this repo is structured
• Prerequisites
• Which setup fits you?
• Set up
  • One command, every agent
  • Claude Code
  • Codex
  • Cursor
  • Copilot / Windsurf / Gemini CLI / other agents
  • Skills only
  • MCP only
• Pin a version
• Verify your setup
• What you can do
• MCP Server
• Troubleshooting
• Repository structure
• Related
• License

What's included

This repo ships two layers that work together, plus plugins that bundle both:

Skills — workflow guidance your agent loads automatically when the conversation matches:

MCP server — live actions in your Tough Tongue AI account. 27 tools over the public API: scenarios, sessions, analytics, organizations, SIP, meeting bots, and collections. Full catalog in MCP.md.

Plugins (Claude Code, Codex, Cursor) — bundle the skills and the MCP server registration in one install.

How the layers relate:

• ttai-agent is the provider-neutral intelligence layer: it turns intent into scoped, capability-aware decisions using the entity handbook and features/mcp adapter guidance.
• Workflow skills add judgment for a job (create, refine, analyze, record a demo).
• With skills only, your agent can advise. With MCP only, it can call tools. With both, it picks the right workflow and completes it. The plugin gives you both in one step.

Need a specific git tag, commit, or a single skill? See INSTALL.md.

How this repo is structured

flowchart TB
  subgraph install["You install one of these"]
    P["Plugin — skills + MCP registration"]
    Sonly["Skills only — guidance, no live tools"]
    Monly["MCP only — 27 ttai tools, no workflows"]
  end
  subgraph layers["Skill layers — load down, don't copy up"]
    L1["Workflows — scenario-maker · session-analyst · browser-demo-builder"]
    L0["ttai-agent — kb (entities, recipes) + features/mcp"]
    L1 --> L0
  end
  P --> layers
  P --> MCP["Hosted MCP https://api.toughtongueai.com/api/public/mcp"]
  Sonly --> layers
  Monly --> MCP

Workflow skills tell the agent when. They load ttai-agent for what exists and how to act. Situation recipes (cold call, sales roleplay, coaching, demo, cascade TTS) live under skills/ttai-agent/kb/scenario-recipes/ — not copied into every workflow skill.

The intelligence layer has no terminal or UI dependency. Coding agents load it before calling MCP; web conversational plugins can pass verified account, workspace, and focused-resource context to the same layer, then execute the result through their server-side adapter.

Prerequisites

Every path needs a Tough Tongue AI account — sign up at app.toughtongueai.com.

That's it for interactive setups. Authentication is OAuth-first: the first time your agent calls a Tough Tongue AI tool, the client opens a browser consent page — approve once and the client stores the token (Claude Code and Codex use your system keychain). No environment variables, no app restarts.

Upgrading from a PAT-based install? After updating the plugin you'll see a one-time OAuth prompt (Claude Code: run /mcp and authenticate; Codex: codex mcp login ttai or approve the automatic prompt). Your existing PAT keeps working for manual/headless configs.

export TTAI_PAT="<your-token>"

Which setup fits you?

Not sure? Use the plugin — it's the least setup. Per-client instructions below. To freeze a git tag, commit, or a single skill, see INSTALL.md.

Set up

One command, every agent

The fastest path. The plugins CLI detects the coding agents on your machine — Claude Code, Cursor, Codex, Grok Build, Kimi Code, GitHub Copilot CLI, VS Code — and installs the skills plus the MCP server into all of them at once:

npx plugins add tough-tongue/toughtongue-skills

Restart your agent, then say "get me started with Tough Tongue AI". The ttai-agent intelligence layer verifies the connection and routes you to the right workflow. On the first tool call your client opens the browser OAuth consent — approve once and you're connected.

This repo is also a standard Agent Plugin (root plugin.json, skills/, mcp.json), so clients that load Agent Plugins natively can point straight at it. Prefer your client's own plugin system? Per-client instructions below.

Claude Code

The plugin bundles the skills and registers the Tough Tongue AI MCP server (.mcp.json) in one install: no separate claude mcp add step needed. On the first tool call, Claude Code opens a browser OAuth consent — approve once and you're connected (re-authenticate anytime with /mcp).

Inside Claude Code:

/plugin marketplace add tough-tongue/toughtongue-skills
/plugin install toughtongue@toughtongue-skills

Or from the terminal:

claude plugin marketplace add tough-tongue/toughtongue-skills
claude plugin install toughtongue@toughtongue-skills

Then say "get me started with Tough Tongue AI" or call ttai:list_organizations. ttai-agent verifies the connection, takes a lightweight inventory, and selects the relevant workflow.

Skills are namespaced after install: invoke them as /toughtongue:scenario-maker, /toughtongue:session-analyst, or /toughtongue:browser-demo-builder; or just describe the task and Claude picks the right skill automatically.