CoreviaFlowCOREVIAFLOWDIGITAL PROJECT COMPANY Leave a request
EN UA RU

Claude API (Anthropic) ↔ Salesforce integration

Claude reads PDF attachments on Salesforce records, drafts notes and follow-up emails and sorts incoming leads, and nothing is written back until a person approves it

Salesforce holds everything about a customer, but much of it is inside attachments nobody has time to read: a tender in PDF, a signed specification, a long request from a new lead. Managers either skim or skip, and notes after calls stay unwritten. The Claude ↔ Salesforce pair gives Claude read access to the records and files you choose, asks it for a summary, a draft or a classification, and shows the result to a person on the record. Only after the person approves does anything change in Salesforce. Below: what exactly moves, which scenarios we cover and which capabilities of both systems we rely on.

What moves between the systems

  • Salesforce → Claude API (Anthropic)A PDF attached to a Salesforce record, such as a tender, specification or contract, goes to Claude together with the record's key fields and a question agreed in advance.
  • Claude API (Anthropic) → SalesforceClaude's summary of the document, with the points your team asked about, appears on the record as a draft note waiting for a person.
  • Salesforce → Claude API (Anthropic)New leads are passed to Claude in batches with their request text, and Claude proposes a category, a priority and a short reason for each.
  • Claude API (Anthropic) → SalesforceProposed lead categories, draft notes and draft follow-up emails are written to Salesforce only when a person accepts them; rejected drafts are simply discarded.

What we connect

  • Together we choose the records and file types Claude may read and the questions it answers for each; nothing outside that list goes to Claude.
  • In Salesforce an administrator sets up access for the pair as an External Client App, which Salesforce now recommends for new connections.
  • The Claude connection runs on your organisation's own Anthropic account; by default Anthropic does not train its models on that data.
  • We describe to Claude the few actions it may propose, such as “set lead category” or “add note”, and our application carries them out only after approval.
  • We test on your real documents and a sample of past leads, compare Claude's proposals with what your team decided, and adjust before launch.

Automation scenarios

01

A forty-page tender

Before the manager postpones reading it and misses a requirement.

After the record gets a summary with deadlines, requirements and risky clauses, which the manager checks against the document and accepts.

02

A pile of new leads on Monday

Before someone reads every request and sorts them by hand.

After each lead has a proposed category and priority with a reason, and the person only confirms or corrects.

03

The follow-up email nobody wrote

Before after a meeting the follow-up waits for a quiet evening.

After a draft based on the record and its latest notes is ready on the record, and the manager edits and sends it.

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.

    Source: privacy.claude.com ↗ checked 2026-09-27
  • Claude answers questions about any text, pictures, charts and tables in the PDF documents you provide; all active models support PDF processing.

    Source: platform.claude.com ↗ checked 2026-09-27
  • 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.

    Source: platform.claude.com ↗ checked 2026-09-27
  • Batch processing: the system creates a batch from the submitted requests and processes it asynchronously, each request independently, for tasks that do not need an instant answer.

    Source: platform.claude.com ↗ checked 2026-09-27
  • REST API gives programmatic access to Salesforce records, metadata and search without using the Salesforce interface itself.

    Source: developer.salesforce.com ↗ checked 2026-09-25
  • Access is authorised through OAuth 2.0 via a Connected App or External Client App; since Spring '26, creating new Connected Apps is restricted and Salesforce recommends External Client Apps instead (existing Connected Apps keep working).

    Source: developer.salesforce.com ↗ checked 2026-09-25

Limits

The limits come from both systems. Claude can make mistakes, so every summary, category and draft is a proposal for a person, and nothing is written to Salesforce without approval. Claude only proposes actions from the short list we describe; our application carries them out. Lead sorting runs in batches, which suits tasks that do not need an instant answer, so results appear after a short wait, not the moment a lead is created. On the Salesforce side, access for new connections is set up as an External Client App, since creating new Connected Apps is now restricted. The pair sends emails only after the manager presses send.

How to start

Three steps from a first message to work on your project.

  1. 01

    Leave a request

    Describe the task in the form — in text or by voice. Budget and timing are optional.

  2. 02

    We review and estimate

    A developer studies the brief, asks about anything unclear and sends an estimate.

  3. 03

    We start

    Once scope and price are agreed, the work begins.

or write to info@coreviaflow.space

Questions

Does our customer data go into training?
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. In addition, only the records and files you chose go to Claude.
Can Claude change a deal or send an email by itself?
No. It proposes; a person approves. Our application writes to Salesforce only the approved result, and only within the actions agreed in advance.
How do we know the lead sorting is good enough?
Before launch we run it on your past leads and compare with the decisions your team made. You see the comparison and decide whether to switch it on.

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.