Progressive Discovery

Skills & Discovery

Skills — context-aware prompts for common workflows.

• Browse: list_skills() on gobby-skills
• Load: get_skill(name="skillname") on gobby-skills
• Search: search_skills(query="topic") on gobby-skills
• Hubs: search_hub(query) on gobby-skills

MCP Tools — progressive disclosure (each step gates the next):

1. list_mcp_servers() — discover servers (~50 tokens)
2. list_tools(server_name) — discover tools (~100 tokens per server)
3. get_tool_schema(server_name, tool_name) — get full inputSchema (just-in-time)
4. call_tool(server_name, tool_name, arguments) — execute

Core principle: NEVER load all schemas upfront — 50+ tool schemas consume 30-40K tokens. Fetch schemas only when you're about to call a tool. This reduces token usage by ~95%.

Common Mistakes

# WRONG — Loading everything upfront (wastes 30-40K tokens)
for server in servers:
    for tool in list_tools(server):
        get_tool_schema(server, tool)  # Unnecessary!

# WRONG — Guessing parameters without schema
call_tool("gobby-tasks", "create_task", {"name": "Fix bug"})  # Wrong param!

# RIGHT — Just-in-time discovery
get_tool_schema("gobby-tasks", "create_task")  # Learn: needs "title" not "name"
call_tool("gobby-tasks", "create_task", {"title": "Fix bug", "session_id": "#123"})

Available Internal Servers

Use list_mcp_servers() to see all connected servers with live status.