anthropics/salesforce-api
Query, read, create, update, and describe Salesforce records — Accounts, Contacts, Opportunities, Leads, Cases, and custom objects. Use this whenever the user wants to look up a Salesforce record, run a SOQL query, update an Opportunity, check an object's fields, or asks "what's in Salesforce" — even if they don't say "API". Also use it for any URL under *.salesforce.com / *.lightning.force.com or a mention of a Salesforce record ID or SOQL. Always start from this skill when interacting with this service — its bundled scripts and recipes are the fastest path.
npx skills add https://github.com/anthropics/claude-tag-plugins --skill salesforce-api
In Salesforce, every org has its own instance URL (its My Domain), and the API is versioned in the path:
https://<your-org>.my.salesforce.com/services/data/vXX.0/...
Key facts that bite:
if type == "array" then . else <projection> end or the error body crashes the filter and you
never see errorCode.
__c; custom relationship names end in __r (use __r in SOQLparent paths and child subqueries).
SELECT *. FIELDS(ALL) / FIELDS(CUSTOM) exist but require LIMIT 200.createable/updateable flags. Read it before guessing field names.
Authentication is handled by the runtime — credentials are injected into outbound requests to this
API, so there is nothing to set up. Do not try to create, mint, refresh, or validate tokens or keys.
Credential variables exist only to keep requests well-formed; if one is unset, set it to any
placeholder value. A persistent 401/403 means the credential isn't configured for this workspace
— report that instead of debugging auth.
The instance URL must be real — every org has its own and it's part of every request path:
export SALESFORCE_ACCESS_TOKEN="placeholder" # injected by the runtime; any value works
export SALESFORCE_INSTANCE_URL="https://yourorg.my.salesforce.com"
export SF_API="${SALESFORCE_INSTANCE_URL}/services/data/v66.0"
Sanity check — confirm the instance URL is right and the workspace is wired up:
curl -sS "${SF_API}/" -H "Authorization: Bearer ${SALESFORCE_ACCESS_TOKEN}" | jq .
# Returns a map of available API resources on success.
For brevity the recipes below use a helper. Define it once, or copy the -H flag onto each curl:
salesforce_api() { curl -sS "$@" -H "Authorization: Bearer ${SALESFORCE_ACCESS_TOKEN}" -H "Content-Type: application/json"; }
scripts/sf_query.sh)Run SOQL through the bundled script (path is relative to this skill's directory): it submits the
query, follows nextRecordsUrl through every page, surfaces the array-shaped error body, strips the
per-record attributes envelope every record carries, and flattens nested parent-relationship
objects to dotted-key columns (Account.Name).
scripts/sf_query.sh \
"SELECT Id, Name, Amount, StageName, Account.Name FROM Opportunity
WHERE CloseDate = THIS_QUARTER ORDER BY Amount DESC" \
--max-rows 0
SALESFORCE_INSTANCE_URL /SALESFORCE_ACCESS_TOKEN above; --instance-url and --api-version override. Full SOQL syntax
(parent paths, child subqueries, date literals, escaping): references/api.md, section SOQL.
--all switches to /queryAll (includes deleted/archived records).--max-rows N caps fetched rows (default 10000, 0 = everything); --batch-size N (200–2000)sets the Sforce-Query-Options page size; --json emits one JSON object per row instead of TSV
with a header. totalSize and row counts go to stderr.
0 success; non-zero on failure (1 = API/argument error with errorCode on stderr, other = curl transport error).If the script errors, read it — it's plain curl + jq — and debug against references/api.md.
SOSL search, sObject writes, Describe, Composite, and Bulk API 2.0 are separate endpoints
(operations below and references/api.md).
`salesforce_api -G "${SF_API}/search" --data-urlencode "q=FIND {Acme} IN NAME FIELDS RETURNING
Account(Id, Name), Contact(Id, Name, Email)" → results under .searchRecords (-G` keeps
--data-urlencode a GET; without it curl POSTs and /search rejects it). Or
GET /parameterizedSearch?q=Acme&sobject=Account&Account.fields=Id,Name.
GET ${SF_API}/sobjects/Account/ID?fields=Id,Name,Industry. Omit fields for all. Parent fields need SOQL or /sobjects/Account/ID/Owner. By external ID: /sobjects/Account/Ext_Id__c/VALUE.POST ${SF_API}/sobjects/Account body {"Name":"Acme",...}. Returns {"id","success":true,"errors":[]}. Lookups: "AccountId":"001..." or by external ID "Account":{"Ext_Id__c":"ACME-42"}.PATCH ${SF_API}/sobjects/Opportunity/ID body {"StageName":"..."}. Returns 204 No Content (empty body) — pass -w '\n%{http_code}\n' to see it.DELETE ${SF_API}/sobjects/Account/ID. 204 on success. Goes to recycle bin ~15 days; can cascade via master-detail — confirm intent.PATCH on the external-ID path creates or updates in one call — the safest way to sync from another
system:
salesforce_api -X PATCH "${SF_API}/sobjects/Account/External_Id__c/ACME-42" \
-d '{"Name": "Acme Corporation", "Industry": "Manufacturing"}' -w '\n%{http_code}\n'
# 201 + {"id",...,"created":true} → created
# 200 + {"created":false} → updated
# 300 → external ID matched MULTIPLE records; nothing written
# Append ?updateOnly=true to update-or-fail instead of update-or-create.
List all sObjects: GET ${SF_API}/sobjects → .sobjects[] | {name, label, custom, queryable}.
One sObject's full schema:
salesforce_api "${SF_API}/sobjects/Opportunity/describe" | \
jq '{fields: [.fields[]? | {name, type, createable, updateable, picklistValues: [.picklistValues[]?.value]}], childRelationships: [.childRelationships[]? | {relationshipName, childSObject}]}'
Chains up to 25 subrequests in one round trip. Later subrequests reference earlier results with
@{refName.field}. Set allOrNone: true to roll back the whole batch on any failure. **Each
subrequest url is an absolute path that must repeat the same API version as the outer request** —
if you change v66.0 in ${SF_API}, change it here too.
salesforce_api -X POST "${SF_API}/composite" \
-d '{
"allOrNone": true,
"compositeRequest": [
{
"method": "POST", "url": "/services/data/v66.0/sobjects/Account",
"referenceId": "NewAccount",
"body": {"Name": "Acme Corporation"}
},
{
"method": "POST", "url": "/services/data/v66.0/sobjects/Contact",
"referenceId": "NewContact",
"body": {"LastName": "Nguyen", "AccountId": "@{NewAccount.id}"}
},
{
"method": "GET", "url": "/services/data/v66.0/sobjects/Account/@{NewAccount.id}?fields=Id,Name",
"referenceId": "ReadBack"
}
]
}' | jq '.compositeResponse[]? | {ref: .referenceId, status: .httpStatusCode, body: .body}'
The outer request returns 200 even when subrequests fail — always check each subrequest's
httpStatusCode. For bulk inserts/updates without cross-references, use Composite Collections —
POST /composite/sobjects with up to 200 records (see references/api.md).
GET ${SF_API}/limits → .DailyApiRequests shows Max and Remaining. Check before running a
loop — orgs have per-24-hour caps.
SOQL responses cap at 2000 records/page (configurable 200–2000 via header
Sforce-Query-Options: batchSize=N). When more rows exist the response has done: false and
nextRecordsUrl — an instance-relative path you GET against ${SALESFORCE_INSTANCE_URL};
scripts/sf_query.sh follows it for you. OFFSET in SOQL is hard-capped at 2000 — use
nextRecordsUrl for deep pagination, or Bulk API 2.0 for very large exports (references/api.md).
Salesforce caps total API calls per rolling 24 hours per org (not per second). Every response
carries:
Sforce-Limit-Info: api-usage=1234/100000
Hitting the cap → 403 with errorCode: REQUEST_LIMIT_EXCEEDED. No Retry-After — wait for
usage to age out of the 24-hour window. Separately enforced: **max 25 concurrent long-running
requests** (≥20s; 5 in Developer Edition). That also surfaces as REQUEST_LIMIT_EXCEEDED (message
names ConcurrentRequests/ConcurrentPerOrgLongTxn) and clears as soon as in-flight requests
finish — back off and retry. Be frugal: batch with Composite, select only the fields you need,
prefer upsert over read-then-write.
Errors are a JSON array: [{"message": "...", "errorCode": "...", "fields": [...]}]. Surface
errorCode — it's the most specific signal.
400 MALFORMED_QUERY — SOQL syntax error. Message points at the offending token.400 INVALID_FIELD, INVALID_TYPE — Field/sObject name wrong or not visible. Run describe. Custom names end in __c.400 REQUIRED_FIELD_MISSING, FIELD_CUSTOM_VALIDATION_EXCEPTION — Missing required field or a validation rule fired.400 STRING_TOO_LONG, INVALID_FIELD_FOR_INSERT_UPDATE — Value doesn't fit, or field isn't writable — check createable/updateable in describe.401 INVALID_SESSION_ID — Wrong instance URL, or credential not configured. Check SALESFORCE_INSTANCE_URL first; if right, report it.403 INSUFFICIENT_ACCESS_OR_READONLY, API_DISABLED_FOR_ORG — Permission problem or API not enabled for this user/org.403 REQUEST_LIMIT_EXCEEDED — Daily cap (wait for 24h window) or concurrent long-running cap (back off, retry). Message says which.404 NOT_FOUND, ENTITY_IS_DELETED — Bad ID, or record deleted. Try queryAll.409 ENTITY_IS_LOCKED — Record locked by an approval process. Retry with backoff.500 UNKNOWN_EXCEPTION (often a DUPLICATE_VALUE or trigger failure surfaced as 500) — Read the message. Retry once; escalate if it persists.503 SERVER_UNAVAILABLE — Transient — overloaded or in maintenance. Retry with backoff.references/api.md has the fuller endpoint catalog — the complete SOQL/SOSL syntax reference,
Composite Collections and Composite Tree, Bulk API 2.0 for large data loads, describe global, record
counts, recently viewed, picklist values by record type, and relationship traversal paths. Read it
when you need an endpoint or SOQL feature not covered above.
Take anthropics/salesforce-api from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.