DevvGwardo

Athena MCP — AI skill for Claude Code

AI community

When Hermes is stuck, it asks Athena.

How to install Athena MCP

This entry records only its repository, not the path inside it, so there is no exact command to give. Open DevvGwardo/athena-mcp and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Athena MCP does

When Hermes is stuck, it asks Athena. An MCP server with a single think tool — a no-tools reasoning sidekick for your AI agent. Backs on Claude subscription or any OpenRouter model.

Alternatives in AI

Zen MCP Server — The power of Claude Code + Gemini / OpenAI / Grok / OpenRouter / Ollama / Custom Model working as one 11.3k ★
Codex Skill — by klaudworks - Enables users to prompt codex from claude code 914 ★
Wax — Shared Single-file memory layer for all your agents, sub mili-second RAG over text, photo and video on Apple S 791 ★

README

athena-mcp

*When Hermes is stuck, it asks Athena.*

An MCP server that gives your tool-using agent a **reasoning sidekick**. One tool: `think`. No side effects — Athena has no tools of her own. She just reasons.

When your main agent (Hermes, Claude Code, Cursor, any MCP-speaking client) hits a hard problem — a subtle bug, an architecture call, a plan that needs critique — it calls `think` and gets back a concise, well-reasoned response. Your agent stays in the driver's seat; Athena is the quiet consultant it turns to when the problem is thornier than its default model handles well.

Inspired by [Codebuff's thinker agent](https://github.com/CodebuffAI/codebuff/blob/main/agents/thinker/thinker.ts), which does exactly this internally: Codebuff's orchestrator spawns `thinker-gpt` with no tools after gathering context, and the thinker's sole job is to think hard and return a brief answer.

Why the separation?

Most agents run on one model for everything. That model is a compromise: fast and cheap enough for hundreds of tool calls, smart enough for most of them. But when it's genuinely stuck, you want a *different* model — a reasoning-heavy one (Claude Opus, GPT-5, Gemini Pro, DeepSeek R1) — without ceding control of the rest of the task. That's what Athena is for.

Cost — reasoning models are expensive to run on every turn. Call them only when needed.
Latency — reasoning models think slowly. Save them for hard problems.
Tool orthogonality — Athena has no tools on purpose. The caller stays in control of side effects.
Model portability — swap reasoning models per call without reconfiguring your whole agent.

Two backends

**`claude-code` (default when the `claude` CLI is on PATH)** — spawns `claude -p` for each call and uses your Anthropic **Pro/Max subscription** OAuth. No API key, no per-token billing. Just counts against your subscription quota. Supports `opus`, `sonnet`, `haiku`, or full Claude model names.

**`openrouter`** — HTTPS