| name | advisor |
| description | When and how to consult the mcp__advisor__consult tool — a stronger reasoning model (Opus) that reviews your context and returns strategic guidance. Apply whenever choosing an approach on a non-trivial task, when stuck, when considering a change of direction, or before declaring a task complete. |
Advisor Tool
You have access to an mcp__advisor__consult tool backed by a stronger
reasoning model (Opus). It reviews your context and provides strategic
guidance in under 100 words as enumerated steps.
When to call
- Before committing to an approach on non-trivial tasks
- When stuck — errors recurring, approach not converging
- When considering a change of approach
- Before declaring a task complete (write/save files first)
When NOT to call
- If you are running as Opus — you are already the strongest model
available, and calling the advisor would be redundant
- For simple, mechanical tasks where the next step is obvious
- Back-to-back without doing meaningful work between calls
- For quick lookups or exploration (read files first, then call)
How to call
Provide rich context in the context parameter:
- What the task is
- What you've learned or read so far
- What approach you're considering or have started
- What's blocking you or what you're uncertain about
The advisor ONLY sees what you put in context. Be thorough — it has
no access to your conversation, your file reads, or your tool history.
Optionally provide a specific question if you have one beyond general
"is my approach right?" guidance.
How to treat the response
Give the advice serious weight. If it conflicts with evidence you have
(a file says X, a test shows Y), surface the conflict in another advisor
call — "I found X, you suggested Y, which should I follow?" — rather than
silently diverging. The advisor may have underweighted your evidence; a
reconcile call is cheaper than committing to the wrong branch.
