Claude Connectors — Drive, Notion, Slack (2026)

Claude Connectors let the Claude app search and read Drive, Notion, and Slack in-conversation. Here's what each does, scope limits, and where they break.

What this covers

Claude Connectors are Anthropic’s official integrations inside the Claude app (claude.ai web and desktop). Once connected, Claude can search and read content from your authorized third-party tools during a conversation, instead of you copy-pasting back and forth. This guide walks through three of the most common workplace tools — Drive, Notion, Slack — and shows what each connector actually does, what to ask, and where the limits are.

What each tool is, and what Claude can do with it

Google Drive

Google Drive is Google’s cloud file storage. The things you typically keep there: Google Docs, Sheets, Slides, PDFs, and uploaded Office files.

What Claude can do via the Drive connector:

  • Search your Drive by keyword or filename.
  • Open and read Google Docs / Sheets / Slides / PDFs (read-only — it will not modify your originals).
  • Compare, summarize, and extract key points across multiple files.
  • Quote specific passages or cells and include file links in its reply.

Good prompts to try:

  • “Find the last month of product weeklies in Drive and pull the biggest risk from each into a table.”
  • “Compare 2026 Q1 OKR and 2026 Q2 OKR — list goals that were added, removed, or rewritten.”
  • “Read the Pricing v3 PDF and explain it to a non-finance teammate in 3 short paragraphs.”

What it cannot do: write back to Drive (no creating or editing files); read folders it was not authorized to; replace a spreadsheet for heavy calculations (it is for analysis and summaries, not number-crunching at scale).

Notion

Notion is a collaborative workspace that combines docs, notes, databases, kanban boards, and project tracking. Its two core building blocks are pages (document-like) and databases (each row is a record, each column is a property).

What Claude can do via the Notion connector:

  • Search authorized pages and databases in your workspace.
  • Read page bodies (including nested subpages) and database row content and properties.
  • Filter database rows by property (e.g. only rows where status=In Progress).
  • Roll up content scattered across multiple pages and databases into a single brief, status update, or weekly report.

Good prompts to try:

  • “From my Notion ‘Engineering Roadmap’ database, list rows where status=Blocked and owner=me, sorted by priority.”
  • “Summarize the last 4 weekly entries in Weekly 1:1 Notes and surface 3 themes that keep coming back.”
  • “Read the Product Spec - Onboarding v4 page and list the ambiguous points an engineer would want clarified.”

What it cannot do: by default it is read-only — it will not create pages or write to databases for you; complex formula properties may not come through completely; the first sync of a very large workspace can be slow.

Slack

Slack is your team’s chat tool. Conversation is organized by channels and DMs, and each message can have a thread and reactions.

What Claude can do via the Slack connector:

  • Search messages within the channels you have access to, by keyword, user, or time range.
  • Read recent channel discussions, entire threads, and pinned messages.
  • Turn a stretch of chat into meeting notes, decisions, action items, or extract commitments and deadlines.
  • Roll up the same topic across multiple channels (e.g. an incident discussed in both #oncall and #eng).

Good prompts to try:

  • “Summarize the last 7 days of #product-launch into three buckets: decided, pending, blocked.”
  • “Find yesterday’s database alert thread in #oncall — list the timeline and the final root cause.”
  • “Search the last 30 days for any message mentioning Vendor X, sorted by date, with who said it and the context.”

What it cannot do: it will not post messages on your behalf (the standard connector is read-only); it cannot read private channels or other people’s DMs you were never in; on very active channels, a too-wide time range will overflow context, so scope by week or by thread.

Who this is for

Claude Pro / Team / Enterprise users with real work content sitting in at least one of Drive, Notion, or Slack. Free plans do not currently include Connectors.

When to reach for it

  • The answer lives in your internal material but is scattered across many files / pages / channels.
  • You want Claude to cite a specific source rather than answer from general knowledge.
  • The task repeats (weekly report, monthly review, project kickoff) and is worth turning into a saved prompt.

When not to

  • A one-off question — it is faster to drag the file into the chat.
  • Your account mixes a lot of unrelated personal content that you cannot wall off with minimum scope.
  • Compliance forbids sending this material to external models. Check with security / legal first.

Before you start

  • Sign in to the Claude app on the account you actually want to connect (a personal Pro account vs. a Team / Enterprise account will authorize different workspaces).
  • Be clear about the task you are trying to solve — not just “let’s hook it up and see.”
  • Have a test sample ready: one Drive file, one Notion page, one Slack channel — to validate on a small surface first.

Step by step

  1. Open connector settings: Claude app → Settings → Connectors → find Google Drive / Notion / Slack.
  2. Click Connect and walk through OAuth: you will be redirected to the provider’s official sign-in. For Drive, pick a Google account; for Notion, pick the workspace and tick the specific pages / databases to share; for Slack, pick the workspace and confirm the requested scopes.
  3. Grant minimum scope: in Notion, do not tick “entire workspace” — select only the pages / databases you need. In Drive, if a folder-scoped option exists, restrict it. Slack does not let you pick channels, but Claude still cannot read channels you were not in to begin with.
  4. Reference content by name in chat: write prompts like “find … in Drive,” “read the Notion page Project A,” “summarize what was discussed in #frontend yesterday.” Claude decides whether to call the connector.
  5. Verify citations: Claude returns filenames / page links / message timestamps — click through and confirm the content was not paraphrased into something different.
  6. Disconnect when done: Settings → Connectors → Disconnect revokes the OAuth grant.

First-run exercise

  1. Pick a low-risk sample: an internal doc, a database view, a non-sensitive channel — not something containing salary or customer data on your very first run.
  2. Run one well-scoped prompt, e.g. “summarize the 5 key points in this doc with quotes.”
  3. Save the result and mark it: usable as-is, needs human editing, or wrong.
  4. Change one variable on the second run — prompt, file, or output format — so you can attribute the difference.

Quality check

  • Did it solve the actual goal, not just produce something longer and more polished?
  • Are the cited filenames, pages, and message timestamps real, and is the content quoted accurately?
  • Were there sources Claude obviously should have pulled in but didn’t? (Often means scope was too narrow, or the keyword missed.)
  • Did the output leak anything sensitive into a place it shouldn’t go — especially notes you’re about to forward externally?

How to reuse this workflow

  • Save successful prompts as templates; next time only swap the filename / channel / time range.
  • Give each connector use case a name (“Drive: weekly rollup,” “Notion: roadmap status,” “Slack: incident retro”) so future-you remembers why it was set up.
  • Re-run a small sample every few weeks because Claude’s connector coverage, UI, and rate limits change.

Authorize (minimum scope) → reference specific sources by name in the prompt → make Claude return links and citations → click through to verify → save the working prompt as a template.

Common mistakes

  • Over-granting scope. Connecting an entire Notion workspace means “find one doc” pulls in unrelated pages and dilutes the answer.
  • Leaving stale connectors enabled. Changed jobs or projects but the old workspace is still connected — both a privacy risk and a source of noisy results.
  • Trusting Claude’s claims without checking. Connector content still gets paraphrased by the model. For anything involving numbers, dates, or decisions, click through to the original.
  • Reading “I don’t have access to that tool” as a model limitation when it is actually a dropped OAuth grant or a per-plan tool toggle (Claude tool unavailable).
  • Time ranges too wide. “Summarize the entire Slack channel history” almost never works well. Scope by week or by thread.

FAQ

Q: Can Claude Connectors modify files in Drive, Notion, or Slack? A: No — connectors are read-only. Claude can search, open, and quote your content, but it cannot create, edit, or delete originals. Anything it produces lives in the Claude conversation until you copy it out yourself.

Q: Why does the Drive connector miss files I know exist? A: Indexing lag and permissions. Newly added files can take a few minutes to surface, and Claude only sees files your Google account has direct access to. Shared-drive permissions inherited via a group sometimes aren’t visible to the connector.

Q: Should I scope queries by time range? A: Yes — “summarize the entire Slack channel” rarely works. Scope by week, by thread, or by sender. Connectors fetch a bounded result set, so a narrow query returns better signal than a wide one.

Q: What does Claude do with my data when a connector is active? A: It reads on-demand in the active conversation; it does not train on connected content by default. Disconnect a connector from Settings if you want to revoke access, and check Anthropic’s privacy docs for the current data handling.

Tags: #Claude #Tutorial