cli

CLI

Five bash scripts, curl and jq, no dependencies. If you can read lib.sh you can call every action in the toolkit without a client library.

Script	Does	Writes?
lib.sh	act <action> <json> helper + require_ready preflight	—
source-to-csv.sh	Search, page, dedupe, drop existing connections, write a CSV	no
engagers-to-list.sh	Post likers and commenters → a named list	no
draft-and-queue.sh	CSV → dry-run invites → queue for human approval	queues only
daily-report.sh	Quota, queue, campaign stats, unread inbox	no

Run

npx linkedin-toolkit-mcp        # pair the extension once
lit serve --http --fake         # drop --fake for a real session

chmod +x *.sh
./daily-report.sh
./source-to-csv.sh "CTO fintech London" ctos.csv 100
./draft-and-queue.sh ctos.csv           # dry run — nothing queued, nothing sent
./draft-and-queue.sh ctos.csv --queue   # queued for approval in the popup

Requires bash, curl, and jq. On Windows, Git Bash or WSL.

lit does most of this in one line

These scripts exist to show the HTTP surface. For day-to-day work the bundled CLI is shorter:

lit status
lit search "CTO fintech London" --source salesnav --count 100 --csv out.csv
lit profile https://www.linkedin.com/in/... --full --json
lit engagers <post-url> --list "Post engagers 8 Sep"
lit invite <profile-url> --note "..."          # queues in Copilot mode
lit campaign create --from sequences/warm-connect.json --list "..."
lit inbox --since 24h --sentiment
lit research input.csv --out ./packs
lit sql "select company, count(*) from profiles group by 1 order by 2 desc limit 20"
lit export --table profiles --csv

lit ships in the same npm package as the MCP server, so npx linkedin-toolkit-mcp already put it on your machine. Full reference: ../../docs/cli.md.

The pattern, in four lines

curl -sS -X POST "$LINKEDIN_TOOLKIT_URL/actions/search.people" \
  -H "authorization: Bearer $LINKEDIN_TOOLKIT_TOKEN" \
  -H 'content-type: application/json' \
  -d '{"keywords":"CTO fintech London","count":50}' | jq '.data.profiles'

Success is { "ok": true, "data": …, "rateLimit": … }; failure is { "ok": false, "error": { "code", "message", "retryAfter", "howToFix" } }. lib.sh's act() unwraps the first and prints the second to stderr before exiting non-zero, which is why none of these scripts contain error handling of their own.

Two habits worth copying

Preflight. require_ready refuses to run if you are not logged in or a challenge is outstanding, and prints your remaining quota and current mode to stderr before doing anything. Every script starts with it. A script that discovers halfway through 80 profiles that the account was challenged has already done the damage.

Dry run first. draft-and-queue.sh with no flag calls outreach.invite with dry_run: true, which returns { "status": "dryRun", "wouldSend": … } and touches nothing. You read what would go out, in full, before any of it is real. Every write action accepts dry_run.

Scheduling

daily-report.sh is read-only and safe to cron:

0 8 * * 1-5  cd ~/linkedin-toolkit/examples/cli && ./daily-report.sh | mail -s "LinkedIn" me@example.com

Do not cron anything that writes. The engine already schedules sends inside your business hours with jittered delays — a cron job on top of it just means two things pacing the same account, and the one you cannot see will win.