How to tell whether Claude is really connected to your Follow Up Boss

An assistant with no working connection to your CRM will still tell you, confidently, that it is connected.

That is not a malfunction. The connector lives inside your assistant, not inside Follow Up Ace, so when you ask Claude whether it can reach Follow Up Boss there is nothing on our side for it to check before replying. “Yes” is the cheapest answer available, and it is the one you usually get.

Three checks settle it properly, and they take about a minute between them. Then there is a fourth reading, which almost nobody takes: the number of tools your assistant is offered. That one is arithmetic, and it answers a question the three checks can’t — whose permissions you are actually looking through.

Why “are you connected?” is the wrong question

A connector can fail in three places that feel identical from the chat window. The authorization can lapse. It can be authorized but switched off in the conversation you happen to be typing in. Or it can be authorized, switched on, and still unable to read your book because your own Follow Up Boss login was never linked to it.

None of them announces itself, and asking the assistant directly doesn’t separate them — which is why our help center says plainly that a general “are you connected?” question is not a reliable test. The checks below are ordered cheapest-first.

Check 1 — Is Follow Up Ace still authorized?

Open your connector list. On claude.ai, Anthropic documents the path as Customize, then Connectors; on Team and Enterprise, owners manage them under Organization settings, then Connectors. Find Follow Up Ace. If it is asking you to sign in, reconnect, or re-authorize, stop here: that is the most common cause we see, and nothing downstream will work until you clear it.

One detail rules out a whole family of wrong guesses. Anthropic’s documentation states that when you add a custom connector, “Claude connects to your remote MCP server from Anthropic’s cloud infrastructure, rather than from your local device.” Your laptop’s network, your VPN and your office firewall are not in the path. If it works on your phone and not your desktop, the difference is in your account settings, not your Wi-Fi.

Check 2 — Is it switched on in this conversation?

Claude turns connectors on per conversation. Anthropic’s guide describes the “+” button at the lower left of the chat, then Connectors, with “toggles allowing you to enable/disable them per conversation.” A connector that is authorized but toggled off in this thread behaves exactly like a broken one, and it is the failure people re-authorize three times without fixing.

Start a fresh conversation, switch Follow Up Ace on in it, and move to check 3 — carrying an old thread forward is how a stale toggle survives a reconnection.

Check 3 — Ask for something only your CRM knows

This is the test that settles it. Paste this, with a real name from your database:

Using Follow Up Ace, look up [a real contact in your Follow Up Boss] and tell me when I last contacted them.

A pass looks like the actual person and an actual date. A fail looks like one of three things: a polite statement that it has no access to your CRM, a description of what it would do if it had access, or an answer you cannot match against a real record. The third is why the prompt names a contact you can check yourself.

If check 3 fails after checks 1 and 2 passed, the connection is live but your own Follow Up Boss login is not linked to it — a two-minute fix, covered in the help center article on connecting Ace to your own Follow Up Boss login.

The tool count tells you whose permissions you’re on

Claude lets you see the individual tools a connector exposes — Anthropic’s guide describes a “Search and tools” menu where you can “disable any tools that aren’t relevant to the current conversation.” Open it, find Follow Up Ace, and count. The number is a direct readout of the seat your connector signed in as, because the connector filters its tool list by caller identity before Claude ever sees it.

Counted against the live assembly on 5 October 2026:

The gaps are not arbitrary. Twenty-one tools are team-wide by definition — per-agent production, response-time leaderboards, the conversion funnel, source ROI, the team cockpit — and an agent seat does not get them, which is the whole 189-to-168 difference. Another twenty-eight are seat-scoped — your own NOW queue, your own daily brief, your own alerts, your own transactions — and an admin who holds no seat has no “my” for them to point at, which is the whole 189-to-161 difference. Eleven of the 189 belong to Ace Trove, the account-wide intelligence layer, which is what the right-hand column removes — a different number per row, because a seat only loses the Trove tools it could see in the first place.

So the count is a diagnosis, not trivia. An agent who expected the team cockpit and finds 168 tools is not broken; they are correctly scoped. A broker seeing 161 instead of 189 has signed in on an identity with no seat attached, and the fix is to reconnect on the Follow Up Boss login their seat is actually on.

The honest limit: these are today’s numbers and they move whenever we ship a tool. Treat the count as a comparison, not a fingerprint — compare your list against a teammate’s on the same day, or compare the shape of it (are the team-wide tools there or not?) rather than memorizing 189. If your client groups or truncates the list rather than laying all of it out, compare the shape and skip the count entirely. And there is only one seat-accurate list anywhere: the one your own assistant shows you. The tool directory on Ace’s Integrations page is a catalogue of everything Ace offers, not a readout of what your connection serves, so counting against it will always look like a shortfall.

Two authorizations, and the agent one is the one you’re reading through

Ace holds two Follow Up Boss authorizations rather than one, and knowing which is which explains most connector confusion. An admin authorizes the account-level one, and it exists for account-level jobs: ingesting webhooks, the nightly scoring sweep that writes fields back into Follow Up Boss, the team-wide analytics.

Every agent then authorizes their own. Everything Ace does for that agent — the connector included — runs on that grant, so Follow Up Boss applies that agent’s permissions and Ace inherits them rather than inventing its own. An agent who hasn’t connected is asked to connect. They are not quietly answered out of the admin’s authorization, because there is no honest way to answer a question about my book with somebody else’s credential.

That design is also why a perfectly configured connector can still decline a read. A correct server URL, a valid authorization and a full tool list say nothing about whether your personal Follow Up Boss connection exists. When it doesn’t, Ace asks you to connect rather than guess — so the prompt that reads like a failure is the system working. The connection article walks through it.

The 60-second checklist

Frequently asked questions

Does Follow Up Ace have a “connected” indicator I can check?

No, and that is a consequence of where the connector lives. The Integrations page inside Ace gives you your server URL, the setup steps for each client and a browsable directory of Ace’s tools by category — but not a seat-filtered list, and not whether your assistant is talking to it right now. That status lives in the assistant, which is why all three checks above happen there.

Why does my teammate see more tools than I do?

Because the tool list is scoped to the seat the connector signed in as. Team-wide reporting tools are advertised to admin and broker seats and withheld from agent seats; the per-agent “my…” tools need a seat to belong to. Both of you are seeing the correct list for who you are.

Does the connector need an Ace Pro seat?

For a seat-holder, yes: the connector is a Pro capability, and the entitlement is re-checked when a call executes rather than trusted from sign-in. Our setup article notes that a Regular seat is turned away before a verification code is even sent, which is why “the code never arrived” is usually a plan question rather than an email one. There is one exception, and the table above already contains it: the seatless-admin row exists because an admin can be admitted without a Pro seat of their own.

Why does ChatGPT only offer search and fetch?

That is the expected shape on the plain connector URL, which serves the two-action contract OpenAI’s Deep Research expects. The full toolset is an explicit opt-in: add ?tools=full to the server URL. Our setup article notes that ChatGPT reads the action list when the connector is added, so refreshing an existing one won’t change it — remove it and add it again. Claude never needs the opt-in: it is served the full toolset by default, still scoped to your seat as described above.

Can an agent borrow the admin’s connection instead of making their own?

No. The account-level authorization drives account-level work; an agent-facing read runs on that agent’s own grant so Follow Up Boss applies their permissions. An agent without one is prompted to connect rather than served someone else’s view of the database.

Where to go next

For the idea rather than the troubleshooting, the Model Context Protocol explainer covers what the standard does, over the protocol’s own remote-server documentation. The copy-paste piece walks through what a confirmed connection is worth. The connector page carries the current setup steps, and the Follow Up Boss guide is the place to start if the CRM itself is new to you.