Ask Claude about your own book.

Seven read-only tools, scoped to your account and nobody else's. Every query runs under the same row-level rules the app itself runs on, so a token reads your book and cannot reach anybody else's.

list_portfolios

Your portfolios and the strategy each one applies.

get_strategy

One portfolio's criteria as the actual thresholds, weights and hard constraints you set.

list_holdings

The tickers you hold or watch, and which portfolio each belongs to.

get_matrix

Every tracked ticker against every portfolio: fit score and verdict.

score_holding

One ticker against one portfolio, broken down per criterion - which rules it meets and which it misses.

list_price_alerts

Your price targets: which are armed, which were reached, and the close that decided it.

get_entitlements

Your plan limits and current usage, so a model can check before spending from the AI pool.

• It never writes. No tool creates a portfolio, tracks a ticker, sets a price target or edits a rule. A model that can add one holding can add fifty, and you would find out from your own book.
• It never touches a broker. No tool reaches your broker connection, the token behind it, or an imported statement. None of them places an order. Claude reads the holdings in your book, however they got there.
• A fit score is not advice. It measures agreement with rules you wrote yourself. It is not a rating of the company, not a prediction, and not a recommendation - and every scored answer carries that sentence with it, so it survives being pasted into a chat.

1. Issue a token. In Monsa, open Settings and find "Read your book from Claude". Give the token a name you will recognise later - "Laptop", "Claude Desktop" - and copy it. It is shown once and stored only as a hash, so if you lose it the answer is to revoke it and issue another.
2. Add it to your client's config. In Claude Desktop that is claude_desktop_config.json, reachable from Settings, then Developer, then Edit Config. Paste your token in place of the placeholder.

   {
     "mcpServers": {
       "monsa": {
         "command": "npx",
         "args": [
           "-y", "mcp-remote",
           "https://app.monsa.ai/api/mcp",
           "--header", "Authorization: Bearer ${MONSA_TOKEN}"
         ],
         "env": { "MONSA_TOKEN": "monsa_mcp_..." }
       }
     }
   }

3. Restart the client. Then ask it something only Monsa knows - "which of my holdings violate my rules right now?" is a good first question, because a wrong answer is obvious.

If you have seen Settings, Connectors, Add custom connector in claude.ai and wondered whether the endpoint goes there: not yet. That dialog takes a URL and OAuth credentials, and it has no field for the token Monsa issues - so a connector added that way cannot authenticate. Adding OAuth is on the list. Until it lands, the local config above is the way in, and it is the one that works in Claude Desktop and Claude Code.

The terminal is where you look at everything at once. This is where you ask one question without opening it. Some that work on the first day:

Drift

"Which of my holdings violate my rules right now, and which rule does each one miss?"

"Anything that was fitting last month and is only borderline now?"

One company, one strategy

"Take NVDA against my Compounders portfolio, criterion by criterion. Where does it fall short?"

"What would have to change about NVDA's numbers for it to fit?"

The rules themselves

"Read back my Wide Moat strategy in plain English - what thresholds did I actually set?"

"Which of my strategies is the strictest, and on what?"

The watchlist

"Anything I am only watching that already fits one of my strategies?"

"Of the tickers I watch, which is closest to fitting Deep Value?"

Price targets and limits

"Which price targets are still armed, and how far is each from the last close?"

"How much of my AI pool is left this month?"

What it will not do is answer "what should I buy?". There is no tool behind that question, and the model has nothing to answer it with except the general knowledge it already had. What Monsa adds is the one thing it could not know: the rules you wrote, and how far each holding has drifted from them.

Every token is listed in Settings with the date it was last used, so the one you set up on a laptop you no longer own is identifiable. Revoking takes effect on the next call - not at some expiry - and the record that the token existed is kept, because "was this real, and when was it last used" is the first question worth asking after anything leaks.