Salesforce is a leading CRM platform integrating sales, service, marketing, and analytics to build customer relationships and drive business growth
What your AI can do hereConnector snapshot 2026-08-23
Connect Salesforce to ChatGPT or Claude with your own Salesforce login or an API key, and your AI can run 219 of its actions. 140 of those only read and change nothing. 24 can delete or send something that cannot be undone, so those need a boundary you set before anyone runs them.
Salesforce also publishes its own MCP server, which is a second way in. Vendor documentation
This describes what Salesforce supports, which is one half of the picture. Where you
start depends on the task and on how much you have handed over before, and no page can
tell you that from a count.
Things you could hand over
Checked 2026-08-22
Candidates, not recommendations. These are things Salesforce supports once connected.
Whether any of them is right for you depends on your own work, which this page knows
nothing about.
Where does the pipeline actually stand?No prior practice needed
SOQL is the query language Salesforce reports are built on, and an assistant can write it for you. You ask in English and get stages, amounts and close dates back as a readout. Querying the CRM changes nothing in it.
What you end up with. A pipeline readout by stage, with amounts and close dates against each.
Start with it bringing you the information and you deciding.
Never without. Ask which fields it queried. Most orgs carry a custom amount or stage field next to the standard one, and the standard one is often the stale one.
Built on: Run SOQL query / Get opportunity / Get all fields for an object
Which deals have gone quiet?Assumes some practice with AI
Filtering on last activity date rather than on stage surfaces the opportunities nobody has touched, which is a different list from the one your stage report shows. It can name the contact on each, so the next step is obvious.
What you end up with. A list of stalled opportunities with the person to contact on each one.
Start with it bringing you the information and you deciding.
Never without. Remember that last activity only counts what somebody logged. A deal that lives in phone calls looks dead here and is not.
Built on: Run SOQL query / Get opportunity / Get contact
Getting an account brief before the callNo prior practice needed
A text search across objects pulls the account, its contacts and the related records in one pass, so you get the history instead of five browser tabs. Every action behind this only reads.
What you end up with. A one page account brief you can read in the two minutes before a call.
Start with it bringing you the information and you deciding.
Never without. Confirm it found the right account. Salesforce orgs accumulate near duplicate account names, and a brief on the wrong one reads perfectly well.
Built on: Execute SOSL search / Get account / Get contact / Get multiple records by IDs
Logging the call outcome onto the recordNo prior practice needed
It updates the opportunity and creates the related activity record. Salesforce returns an empty success on an update, so the honest version of this reads the record back afterward and shows you what it now says.
What you end up with. An updated opportunity, with the new values read back for you to confirm.
Start with it doing the work and checking with you.
Never without. Require a read back after every write. An update returns nothing on success, so without that second call you have no evidence the change landed.
Built on: Update opportunity / Get opportunity / Create SObject record
Can it enter the leads from a list I collected?Assumes some practice with AI
It creates them one at a time, or up to two hundred in a single call. Your org's validation and duplicate rules still apply, so a correctly formed lead can be rejected for a reason that lives in your Salesforce rather than in your data.
What you end up with. The leads in Salesforce, plus a list of the ones your org rejected and why.
Start with it doing the work and checking with you.
Never without. Run the first five before the rest and read the errors. A rejection names the failing field, and that field usually points at a rule nobody told you about.
Built on: Create lead / Create records using sObject Collections / Run SOQL query
Standing up a campaign and getting the right people into itAssumes some practice with AI
Campaigns are creatable and updatable through the connection, and leads and contacts can be added to one individually as campaign members. The assistant can build the target list with a query first, so the membership comes from a rule you stated rather than from a list somebody pasted. Adding a member fails outright when the person is already in the campaign, so it can check membership before it writes.
What you end up with. A campaign in Salesforce with its dates and type set, and the members you queried for added to it.
Start with it doing the work and checking with you.
Never without. Check what your org fires on campaign membership before it adds anybody. In most orgs a new campaign member is what starts a nurture journey or a send, so a query that pulls twenty extra people is twenty emails you did not write.
Built on: Create campaign / Update campaign / Get campaign / List campaigns / Add contact to campaign / Add lead to campaign / Run SOQL query
Updating contacts in bulk without creating duplicatesAssumes you have done this kind of handover before
Upsert matches each incoming record on an external ID field, updates the match, and creates the record where there is none. All of it rests on that ID being genuinely unique in your data.
What you end up with. Contacts brought up to date in one pass, with no second copies created.
Start with it doing the work and checking with you.
Never without. Prove the external ID is unique before you upsert on it. A repeated value updates the wrong record and a blank one creates a new one, and both report success.
Built on: Upsert records using external ID / Run SOQL query / Get contact
Clearing out the junk leads without losing a real oneAssumes you have done this kind of handover before
The connector documents lead deletion as permanent. The workable shape is three steps: a query that lists the candidates, a person who approves that exact list, and a delete that runs against the approved IDs and nothing else.
What you end up with. A cleaner lead table and a saved list of everything that was removed.
Start with it doing the work and checking with you.
Never without. Save the query results and approve the IDs as a fixed list, never as a filter the delete re-runs for itself, because this deletion is documented as permanent and cannot be undone through the API.
Built on: Run SOQL query / Get lead / Delete lead
What we could not establish
Whether any create or update succeeds depends on your org's validation rules, duplicate rules and field level permissions, none of which can be read from outside the org. The connector's own descriptions note that correctly formatted requests are rejected for these reasons.
Contact deletion is exposed and no candidate here uses it, because deleting a duplicate contact is only the last step of a merge. Salesforce's own description says activities and opportunities lose the reference rather than moving to the surviving record, and reparenting them means writes to those objects that were not worked through here.
File upload and file content retrieval are exposed and are not covered here, so putting a signed document on an opportunity was not established as a candidate.
Several deletion and campaign actions appear twice in the catalog, once current and once marked deprecated. The candidates here cite only the current ones.
No time saving is claimed for any candidate on this page, because none has been measured.
BUILT-IN AI Ships AI features you can turn on yourself, with no developer and no new purchase.
7 event triggers. This app can fire automations on its own. One of 34 in this directory that can.
Capabilities
READPull records, messages, and content out on demand.
WRITECreate and update records from an instruction.
BULK-OPSRun one instruction across many records at once.
ADMIN-PROVISIONINGManage users, access, and configuration.