using-agentic-mcp

Using agentic-mcp

Overview

Use this skill when the goal is a working outcome: get a real answer from the provider already installed on the machine, compare multiple providers on one prompt, or configure one MCP client entry instead of wiring providers separately. Start by discovering providers, prove one provider works with ask_<provider>, then use ask_all only for deliberate comparison.

When to Use

• You need to get to a first successful provider answer through agentic-mcp.
• You need to compare multiple providers on the same prompt without bespoke glue.
• You need one repeatable workflow for MCP tools and CLI fallback.

Do not use this for general skill authoring. Use skills-writing for that.

Core Workflow

1. If the environment supports skills, load its "using skills" guidance first.
2. Run minimal onboarding first:
   • npx agentic-mcp init
3. Discover what is detected:
   • list_providers
   • ping_<provider> only when you need a limited proof check
   • help_<provider> only when you need capabilities or CLI details
4. Prove real usage before claiming readiness:
   • use prove for the fastest first real answer
   • use ask_<provider> for focused single-provider work after the first proof
5. Configure MCP client integration only after the first proof when you need it:
   • npx agentic-mcp setup --client claude-code --yes
6. Compare providers deliberately:
   • use ask_all only when comparison itself is the goal
7. If MCP calls are unavailable, use CLI with the same sequence.
8. Report configured files, detected providers, what remains unproven, and whether a real ask succeeded.

Prompt Templates

Setup Prompt

Set up agentic-mcp for me end-to-end and do not stop early.

Execution requirements:
1) If your environment supports skills, load the equivalent of a "using skills" workflow first.
2) Run minimal onboarding first with: npx agentic-mcp init
3) Discover providers with: list_providers
4) Run ping_claude only if you need a limited proof check.
5) Prove real usage first with: npx agentic-mcp prove claude
6) If Claude Code setup is part of the goal, then configure MCP client integration with: npx agentic-mcp setup --client claude-code --yes
7) After setup, get a focused real answer with: ask_claude "Reply with OK and your provider name."
8) Use help_claude only if you need capabilities or CLI details.
9) If any check fails, fix root cause and rerun checks.
10) Return a concise report with configured files, detected providers, what remains unproven, and whether the real ask succeeded.
11) Ask if any other providers should be set up too; if yes, configure and verify them the same way.