Mention the assistant in a Slack channel and Claude answers in the thread, reads the attached PDF and prepares draft replies that a person approves with a button
In most teams the questions are already in Slack: what does this contract say, how do we answer this customer, where is the summary of that document. Someone copies text into a separate AI window, pastes the answer back and loses the thread. The Claude ↔ Slack pair puts Claude where the conversation already is: it answers when mentioned, reads PDF files shared in the thread and prepares draft replies. Anything that leaves the channel or changes data waits for a person to press a button. Below: what exactly moves, which scenarios we cover and which capabilities of both systems we rely on.
Slack → Claude API (Anthropic)When someone mentions the assistant in a channel it was added to, Slack notifies the pair, which checks that the request really came from Slack and passes the question with the thread to Claude.
Slack → Claude API (Anthropic)A PDF shared in the thread, such as a contract, price list or report, goes to Claude together with the question, including its tables and charts.
Claude API (Anthropic) → SlackClaude's answer comes back into the same thread, so the discussion stays in one place and everyone sees what was asked and answered.
Claude API (Anthropic) → SlackWhen a request needs an action, for example a reply to a customer, Claude prepares a draft and the message carries “Approve” and “Edit” buttons; nothing is sent until a person presses one.
What we connect
We agree which channels the assistant works in and what it may do there: answer, read files, prepare drafts.
A Slack workspace administrator installs the pair's app and adds it to those channels only; Slack gives the app a secret used to confirm every request comes from Slack.
The Claude connection runs on your organisation's own Anthropic account; by default Anthropic does not train its models on that data.
We write the assistant's instructions with you: tone, what it must not answer, and when it should hand the question to a person.
We start in one test channel, check answers on your real documents, and only then add the working channels.
Automation scenarios
01
“What does this contract say about penalties?”
Beforesomeone downloads the PDF, reads twenty pages and writes a summary.
Afterthe file is shared in the thread, the assistant is mentioned, and the answer with the relevant clauses arrives in the same thread for a person to check.
02
A reply to a difficult customer
Beforethe manager drafts a reply alone and asks colleagues for comments.
AfterClaude prepares a draft in the thread, the manager edits it or presses “Approve”, and the team sees the final version.
03
A long thread nobody can follow
Beforesomeone returning from leave scrolls through a hundred messages.
Afterthey ask the assistant in the thread and get a short summary of what was decided and what is still open.
How it works technically
For those who want to check: each point links to the official documentation page it comes from.
By default Anthropic does not use inputs or outputs from its commercial products, including the platform connection, to train its models, unless the account owner explicitly allows it.
Tool use: Claude decides when to call a function described by the client based on the request and the function's description, and returns a structured call that the client's application executes.
The limits come from both systems. The assistant sees only the channels it was added to and only the messages where it is mentioned or the threads it is asked about; it does not read the whole workspace. Claude can make mistakes, so answers about contracts, money or customers are a draft for a person, not a decision. Claude can decide to call one of the actions we describe for it, but our application carries the action out, and only after a person approves it. Slack signs each request to the app, and a request without a valid signature is ignored.
How to start
Three steps from a first message to work on your project.
01
Leave a request
Describe the task in the form — in text or by voice. Budget and timing are optional.
02
We review and estimate
A developer studies the brief, asks about anything unclear and sends an estimate.
By default Anthropic does not use data from its commercial products, including the platform connection, to train its models unless the account owner explicitly allows it.
Can the assistant send something to a customer by itself?
No. It prepares a draft in the thread, and only a person's press on “Approve” sends it or changes data.
Does it read scanned documents?
Claude works with the text, pictures, charts and tables in PDF files. How well it reads a particular scan we check on your real documents in the test channel before launch.
Tell us what needs to be connected
Describe the task in text or a voice message. We look at your setup and reply with a plan.