Claude Connectors: What They Are, How to Add Them, and Which to Set Up First

Ask Claude why cost per lead jumped last week and, without connectors,

it can only work with whatever you paste into the chat. Connect your

CRM, your Google Drive, and your ad accounts, and the same question gets

answered from your own data, with the option to act on it. That second

part is where the setup decisions sit, because a connector that can send

an email or change a budget deserves more thought than one that only

reads.

This guide covers what Claude connectors are, how they work, how to add

one, which to set up first for marketing and paid media, and what

changes when a connector can spend money. Menus and plan limits below

were checked against Anthropic's docs and the live interface on October

3, 2026.

What Are Claude Connectors?

**Claude connectors link Claude to the apps you already use, such as

Google Drive, Slack, or HubSpot, so it can read your data and take

actions there.** Claude inherits your permissions in each app, so it

only reaches what your own account can reach. Each connector is an MCP

server built on the Model Context Protocol, an open standard.

[Anthropic's connectors

guide](https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities)

puts the permission rule plainly: if someone can't access a file,

channel, or record in the source system, the connector can't reach it

from Claude either. And because MCP is the standard underneath, "Claude

MCP connectors" and "Claude connectors" usually mean the same thing.

There's one naming trap worth clearing up early, though. In ad tech,

"connector" usually means an ETL or sync connector that copies data

between systems on a schedule, which is the pattern we argue against in

[multi-platform campaign management without

connectors](https://syntermedia.ai/blog/multi-platform-without-connectors).

A Claude connector is a different animal: Claude calls its tools during

the conversation, at the moment your question needs them.

Connectors come in three kinds, and the difference matters more for

trust than for setup:

Kind Where it comes Reviewed by Works in

from Anthropic

• ----------------- ----------------- ----------------- -----------------

Directory Listed in Verified Web, Desktop,

connector Anthropic's connectors are mobile, Cowork,

Connectors tested; Community Claude Code

Directory connectors are

screened but not

reviewed in depth

Custom connector You add a remote No Web, Desktop,

MCP server by its Cowork, Claude

URL Code

Desktop extension A local MCP Depends on the Claude Desktop

server installed source only (Claude Code

in Claude Desktop needs the server

configured

separately)

Directory and custom connectors run on [the same MCP

infrastructure](https://claude.com/docs/connectors/building/directory-vs-custom),

so a listing changes who reviewed the server and how you find it, not

what its tools can do. The labels themselves are worth reading

carefully: Anthropic's [connector verification

page](https://claude.com/docs/connectors/verification) says

a Verified label isn't a security audit, and that a Community connector

was screened before listing but not reviewed in depth.

How Claude Connectors Work

When you ask for something that maps to one of a connector's tools,

Claude calls that tool, and the MCP server, which sits between Claude

and the app, carries out the read or the action on your behalf.

Anthropic's [MCP connector

documentation](https://platform.claude.com/docs/en/agents-and-tools/mcp-connector)

draws the line with an example: "how do Notion databases work?" gets

answered directly, while "what's in my Projects database?" triggers

the tool.

For remote connectors, that call doesn't leave from your laptop.

**Custom connectors are reached from Anthropic's cloud, even when

you're using Claude Desktop or Claude Cowork**, so a server on a

private network or behind a VPN won't connect unless it allowlists

Anthropic's IP ranges, as the [custom connector

guide](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp)

explains. Enterprise organizations can also request access to the [MCP

tunnels](https://claude.com/docs/connectors/mcp-tunnels/overview)

research preview, which reaches private servers over an outbound-only

connection instead. For an ad platform, that middle layer is where

controls such as spend caps can sit, before anything reaches the

platform's own API.

Remote Connectors vs. Desktop Extensions

Most connectors are remote, which is the default and works everywhere

you use Claude. A desktop extension is the exception: a local MCP

server, packaged as an MCP Bundle (.mcpb), that runs on your own machine

inside Claude Desktop. Pick one when the thing you want Claude to reach

lives on your computer (local files, a database on localhost, a desktop

app) and has no cloud version. You install it from **Settings \>

Extensions** in Claude Desktop by dragging in the .mcpb file, and it

won't follow you to the web or mobile apps.

Interactive Connectors (the "Interactive" Badge)

Some [interactive

connectors](https://support.claude.com/en/articles/13454812-use-interactive-connectors-in-claude)

render a live interface in the chat instead of returning text: a compact

inline card for a quick confirmation, or a full-screen view for

dashboards and boards. The directory marks them with an Interactive

badge, and the current list includes Amplitude, Asana, Box, Canva, Clay,

Figma, Hex, and Slack. You can click through them without leaving

Claude; they use the same permissions you granted when you connected the

app, and Anthropic states plainly that they can't make purchases or

financial transactions.

How Claude Decides When to Use a Connected App

Once an app is connected, Claude can bring it into a conversation on its

own when it fits what you're asking, and when two connected apps could

both help, it shows both and lets you choose. Directory connectors can

also show up as Suggested Connectors, recommended in the chat when one

fits your task. Custom connectors are never suggested, so for those,

check that the toggle is on under + \> Connectors for the

conversation, or name the connector in your prompt.

Every connector also takes up room in the conversation. The [Tool

access

setting](https://support.claude.com/en/articles/13730515-manage-claude-s-tool-access)

has three modes: Auto (the default), Always available, and **On

demand**, which Anthropic recommends once you have ten or more

connectors active.

What Data Can Claude Access Through Connectors?

Claude can access whatever your account can access in the connected app,

and nothing more. The exception is a custom connector set up with a

shared credential, such as an API key an admin entered once for the

whole team. That connector reaches whatever the credential can reach,

for everyone who uses it.

**The bigger question is usually what Claude can change, not what it can

see.** Connector tools are grouped into read-only tools and write/delete

tools, and each group or individual tool can be set to Always allow,

Needs approval, or Blocked. When we opened the Gmail

connector's settings on October 3, 2026, it listed six read-only tools

and more than 20 write/delete tools, including Send email message,

Forward email, and Move thread to Trash. OWASP describes this

exact mismatch in its [Excessive Agency

entry](https://genai.owasp.org/llmrisk/llm062025-excessive-agency/):

you need read access, and the extension you installed can also modify

and delete. Before you grant write access to any connector, read its

tool list on the directory page or in the vendor's docs. For our own

server, the [Synter MCP tool

reference](https://docs.syntermedia.ai/mcp/tools) lists all

141 tools and labels each one read-only or write.

The other practical limit is coverage. Anthropic's [Google Workspace

connector

guide](https://support.claude.com/en/articles/10166901-use-google-workspace-connectors)

says Claude retrieves the minimum information needed to answer your

question. Its [Drive connector

documentation](https://claude.com/docs/connectors/google/drive) adds

that a very large file can arrive incomplete, and that the Drive

connector doesn't read comments or suggested edits. So when an answer

depends on completeness, such as "every client email about the Q3

budget," ask Claude to list what it searched and which items it

actually read before you trust the summary.

How to Add a Connector to Claude

Connectors live under Customize \> Connectors. If an older guide

sends you to claude.ai/settings/connectors, you'll land on a note that

connectors have moved to Customize. From there, you can add a listed

connector in a few clicks, or add any remote MCP server by URL.

From the Connectors Directory

1. Go to Customize \> Connectors. In Claude Desktop, select

Customize in the sidebar, then Connectors.

1. Select Discover to browse, or type a name into **Search

connectors**.

1. Open the connector's page, check its Verified or Community label

and its tool list, then select Connect to Claude.

1. Sign in to the service and approve the access Claude asks for. On

claude.ai, the sign-in page opens in the same tab; in Claude

Desktop, it opens in your browser.

1. Back on the Connectors page, confirm the status reads Connected,

then turn the connector on in a chat from + \> Connectors.

Add a Custom Connector With a Remote MCP URL

Here's how to add a custom connector in Claude for a service that

isn't in the directory, or for a server your own team built. Custom

connectors work on Free, Pro, Max, Team, and Enterprise plans, with

Free accounts limited to one custom connector. On an individual Pro

or Max plan, the flow looked like this when we checked it in the live

interface:

1. Go to Customize \> Connectors and select + Add, then **Add

custom connector**.

1. In Name, type the name you want in your connectors list.

1. In MCP server URL, paste the server's HTTPS address, then

select Continue.

1. Review the authentication settings Claude detected (Sign in now,

Sign in when needed, or No sign in, plus the OAuth client

option) and change them only if the server's documentation says to.

Then select Add.

1. Click Connect and finish the sign-in in your browser, then

enable the connector in a chat from + \> Connectors.

Using our own server as the worked example, the two values are the name

of Synter and this URL:

https://mcp.synterai.com

Synter signs you in through browser OAuth, so there's no API key to

paste into Claude. [Synter's own setup

guide](https://syntermedia.ai/installation-instructions) also includes a

read-only first prompt that confirms the right workspace and ad account

before anything else happens.

If you're the one publishing a server, Anthropic supports an install

link that opens this dialog with the name and URL already filled in

(https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=NAME&connectorUrl=ENCODED_URL).

The user still reviews and confirms everything, so it saves typing

without skipping any checks.

Add a Remote MCP Server in Claude Code

Claude Code adds remote servers from the terminal with claude mcp add

\--transport http \<name\> \<url\>, as covered in the [Claude Code MCP

documentation](https://code.claude.com/docs/en/mcp). The

\--scope user flag makes the server available across all your projects,

and /mcp inside Claude Code handles the OAuth sign-in. For Synter, that

looks like this:

claude mcp add \--transport http \--scope user synter-ads

https://mcp.synterai.com

Then start Claude Code, open /mcp, select synter-ads, and complete the

sign-in in your browser. If you've already added a connector in

claude.ai and you sign in to Claude Code with the same account, it shows

up there without this step.

Team and Enterprise: Owner Enablement and Tool Permissions

On Team and Enterprise plans, an Owner or Primary Owner enables each

directory connector under Organization settings \> Connectors

(Browse connectors, then Add to your team). That makes the

connector available, but each person still signs in with their own

account before Claude can use it. On Team plans, members without

permission see a Request button that notifies the Owners. Custom

connectors follow the same pattern: only Owners and Primary Owners can

add them to a Team organization (Enterprise also allows a custom role

with library management), and members then connect individually.

Owners can also restrict what a connector does across the organization,

for example, letting Claude read Google Drive files while blocking it

from creating or editing documents. Individual users can't override

that setting, and it never grants more than the source system already

allows.

Best Claude Connectors by Use Case

The best Claude connectors are the ones that hold the context your

questions depend on, so this list is organized around the work a

marketing or paid media team actually does. The read or write column

reflects the tools each connector's directory listing showed on October

3, 2026, not your own permission settings.

Use case Connector What Claude can Read or write

do

• ----------------- ----------------- ----------------- -----------------

Work context Google Drive Search and read Read and write

Docs, Sheets,

Slides and PDFs;

create, update

and share files

Work context Gmail Search and read Read and write

threads; draft,

reply, send,

label and trash

email

Work context Google Calendar Check Read and write

availability and

suggest times;

create, update

and delete events

Work context Slack Search channels Read and write

and threads; send

messages and

create canvases

Work context Notion Search and fetch Read and write

pages; create and

update pages and

databases

Sales and CRM HubSpot Query CRM records Read and write

and campaign

analytics; manage

CRM objects,

landing pages and

marketing emails

Sales and CRM Salesforce (beta) Discover and Read and write

query org data;

run actions in

Salesforce

Marketing and SEO Semrush Keyword, Mostly read

backlink,

competitor, and

paid search

research; site

audits

Marketing and SEO Ahrefs Keyword research, Mostly read

backlink analysis

and AI search

visibility data

Marketing and SEO Canva Search, generate, Read and write

autofill and

export designs

Paid media and Windsor.ai Query blended Read in the

analytics marketing data listing; Windsor

from 325+ data documents write

sources actions for some

ad platforms

Paid media and TikTok for TikTok Ads Read and write

analytics Business reporting,

account and

audience lookups,

ad creation

Paid media and Synter (custom Reporting and Read and write

analytics connector) campaign changes

across 27 ad

platforms, with

approval modes

and spend caps

Automation Zapier Run actions in Depends on the

the apps you've actions you

enabled in Zapier enable

If you're starting from zero, **connect Google Drive, your email, and

your CRM first**. Those three answer most "what happened with this

account?" questions, and their read tools are useful on day one.

Engineers on the team may want the GitHub Integration too, which is

available on every plan, including Free, and reads repository files

rather than pull requests or issues.

Paid media is where the list gets crowded. A search for "ads" in the

directory on October 3, 2026 returned dozens of results, and most

carried the Community label, meaning Anthropic screened them but

didn't review them in depth. Check the label, the builder, and the tool

list before you connect anything to an account that spends money,

because a Community connector gets the same access as any other once you

approve it.

Connecting Claude to Your Ad Accounts

A connector that drafts the wrong email costs you an awkward follow-up.

A connector that raises the wrong budget costs you real money, and by

the time anyone checks, the spend has already happened. **When a

connector can spend money, the approval step has to live somewhere that

understands budgets, not only in a chat prompt. Claude's Needs

approval** setting asks before each tool call, which is a good start,

but it can't know that a campaign is already at its monthly cap. It

also doesn't cover every mode: in Research, Claude can call connector

tools automatically without further approval, which is why Anthropic's

guidance is to disable write tools when you use Research with custom

connectors.

The directory already has credible options for part of this job.

Windsor.ai is listed there and pulls marketing data from more than 325

sources, which makes it a reasonable pick if blended reporting across

ad, analytics, and ecommerce data is the main need. Windsor's own

documentation now lists write actions for five ad platforms, including

Google Ads and Meta Ads. TikTok's own TikTok for Business connector

covers TikTok Ads, including ad creation, if that's the one platform

you run.

We built [Synter's Universal Ads

MCP](https://syntermedia.ai/mcp) for teams running several

platforms who want one governed path to all of them. Synter isn't in

Claude's Connectors Directory as of October 3, 2026, so you add it as a

custom connector with the steps above. One connection covers 27 ad

platforms, and the actions available still vary by platform API and

account permissions. Here's what changes compared with a reporting-only

connector.

Connecting Claude doesn't connect your ad accounts. Authorizing

Claude to use Synter and authorizing Synter to reach each ad platform

are separate OAuth steps, and installing the connection doesn't

authorize any spend. You can add the connector on a Monday and still

have nothing able to touch a campaign until someone connects an ad

account on purpose.

Approval mode is a setting, not a habit. Each workspace or campaign

runs either in review-required mode, where a person approves launches

and edits, or in auto-pilot within guardrails. That choice lives in

Synter rather than in the chat, so it applies no matter which AI client

or teammate sends the request.

Caps are checked before the platform sees the call. Hard spend caps

are enforced before any API request reaches the ad platform, which means

a prompt that asks for a bigger budget fails at the cap instead of at

the next invoice.

Changes are logged and reversible. Every change records the old and

new values plus the agent's rationale, and [rollback covers the last

10 changes per

entity](https://syntermedia.ai/security-governance). New

Meta and Reddit ads created through our tools also default to paused, so

someone reviews them before they spend.

Whichever connector you choose, start with a read-only report on one

account, keep write tools on Needs approval in Claude, and route

each ad account through a single governed connection rather than several

overlapping ones.

Limits and Safety Tips for Claude Connectors

Plan limits are modest. [Claude's pricing

page](https://claude.com/pricing) includes connectors on

the Free, Pro, and Max plans, and Free accounts can add one custom

connector. Research, which can call your connectors on its own, is a

paid plan feature. The service you connect may still need its own paid

account. HubSpot's connector, for example, needs a paid Claude plan

(Pro, Max, Team or Enterprise) as well as a HubSpot account.

A few habits prevent most connector problems. **Connect only servers you

trust.** A malicious MCP server can hide instructions that try to make

Claude do something you didn't ask for, which is the risk OWASP lists

first in its Top 10 for LLM applications: [prompt

injection](https://genai.owasp.org/llmrisk/llm01-prompt-injection/).

Keep write tools on Needs approval until you've watched how Claude

uses a connector for a while, and block the tools you'll never need.

Turn off connectors you aren't using in a given chat, since they

take up room in the conversation and fewer active tools means fewer

surprises.

For paid media specifically, put the approval step at the system level

as well as in Claude, so a misread prompt can't spend past a limit

someone set deliberately.

Connectors give Claude your access, not more access. Start read-only,

add write access tool by tool, and give anything that can change a

budget an approval step that knows what the budget is. If you want to

try that on your own ad accounts, add Synter as a custom connector, run

the read-only first prompt, and decide what to allow from there. Our

Solo plan, built for developers connecting agents, is \$20 a month, and

brand workspaces for teams run on Scale at \$500 a month per workspace.

The Universal Ads MCP page linked above currently applies a \$100 credit

voucher with code MCP100.

Contact our team to

discuss connecting Claude to your ad accounts.

FAQ

What connectors does Claude have?

Claude's Connectors Directory lists connectors built by Anthropic and

by third parties, covering work apps like Google Drive, Gmail, Slack and

Notion, CRMs like HubSpot and Salesforce, and marketing tools like

Semrush, Ahrefs and Canva. You can also add any remote MCP server as a

custom connector, whether or not it's listed.

Does Claude have connectors like ChatGPT?

Yes, ChatGPT supports remote MCP servers too, through [developer-mode

apps](https://developers.openai.com/api/docs/guides/developer-mode),

so a server like Synter's can be added to either assistant.

Does Claude Pro allow connectors?

Yes, Pro includes directory connectors and custom connectors (the

one-custom-connector limit applies only to Free), and it also includes

Research.

Are Claude connectors free?

Connectors are available on the Free plan, which is limited to one

custom connector. Some connected services need a paid account of their

own, and that's billed by the provider.

Can Claude connectors take actions or only read?

They can do both, depending on the connector: many expose write tools

that create, modify, or delete data, and you decide per tool whether

Claude can use them freely, needs your approval, or is blocked.