Skills
TARX skills are the AI capabilities exposed via MCP tools. Every skill works with Claude Desktop, Claude.ai, and any MCP client.
Available Tools (11)
Status & Health
| Tool | Description |
|---|---|
tarx_status | System health — all ports, daemon, model, uptime |
tarx_cognitive_state | Focus depth, fatigue, recommended interaction style |
Memory
| Tool | Description |
|---|---|
tarx_remember | Store a memory, decision, or note permanently |
tarx_recall | Semantic search over stored memories |
tarx_save_conversation | Persist a conversation summary with key decisions |
Spaces & Files
| Tool | Description |
|---|---|
tarx_list_spaces | List all Spaces (projects) |
tarx_list_files | List files indexed in a Space |
tarx_project_context | Load full project context — files, decisions, activity |
tarx_recent_activity | What was the user working on recently |
Delegation
| Tool | Description |
|---|---|
tarx_delegate | Queue a task for Claude Code to execute |
tarx_pending_tasks | List tasks waiting in the delegation queue |
Tool Annotations
Every tool includes MCP annotations for client optimization:
- readOnlyHint: Tool doesn’t modify state (safe to call speculatively)
- idempotentHint: Calling multiple times produces same result
- openWorldHint: Tool accesses external systems
- destructiveHint: Tool deletes or overwrites data
Client Profiles
TARX detects your MCP client and adjusts tool descriptions:
- Minimal (Claude, Cursor, Windsurf): Short descriptions to save context window
- Rich (custom clients): Detailed descriptions with args, return types, usage hints
OpenAI Function Calling
For OpenAI-compatible models, TARX exports tools in function calling format:
curl https://mcp.tarx.com/openai-toolsThis returns all 11 tools with rich descriptions and JSON Schema parameters.
Pass the result directly to tools in your chat completion request.
Last updated on