gtm-ai-cli

name	gtm-ai-cli
description	This skill should be used when searching for or enriching companies and contacts, surfacing buyer intent signals, fetching real-time business events (scoops) or company news, reading or updating GTM context (offerings, ICPs, personas, competitors), looking up reference values (industries, metros, intent topics, tech products), or doing any ZoomInfo / GTM data lookup from the terminal. Activates when the user asks to "find companies", "search contacts", "enrich a contact", "find lookalike companies", "research a company / person", "get intent signals", "find scoops", "fetch company news", "show our GTM context", "look up industry codes", or any task involving ZoomInfo data lookup from the terminal.
version	0.1.0

GTM AI CLI Skill

Use the gtm CLI to drive ZoomInfo's GTM AI MCP from the terminal: search and enrich companies and contacts, surface intent signals, find real-time business events (scoops), pull news, read/update your organization's GTM context, and look up reference values. Output defaults to JSON for jq piping; use -f, --format to switch to jsonl, csv, yaml, or table.

Authentication

Before running any command, confirm the user is authenticated:

gtm auth whoami

If not logged in:

gtm auth login

To force a fresh OAuth registration (rare — only needed if the saved client_id is stale):

rm ~/.config/gtm-ai/client_id ~/.config/gtm-ai/credentials
gtm auth login

The lookup workflow — read this first

ZoomInfo's search endpoints reject free-text industry/metro/topic strings with HTTP 422. Always look up the canonical ID first, then pass it to the search filters. Treat gtm lookup as a required first step for any filtered search.

# Industries containing "software"
gtm lookup --field industries --fuzzy software -f table

# Management levels (canonical names: Board Member, C Level Exec, VP Level Exec, Director, Manager, Non Manager)
gtm lookup --field management-levels -f table

# Intent topics
gtm lookup --field intent-topics --fuzzy cloud -f table

# Tech products require TWO lookups: vendor first, then products under that vendor
gtm lookup --field tech-vendors --fuzzy hubspot
gtm lookup --field tech-products --vendor "HubSpot, Inc"

# Metro regions (US + Canada only)
gtm lookup --field metro-regions --fuzzy "san francisco"

Valid --field values: industries, metro-regions, states, countries, continents, management-levels, departments, job-functions, employee-count, revenue-ranges, company-types, company-rankings, intent-topics, scoop-topics, scoop-types, scoop-departments, news-categories, naics-codes, sic-codes, tech-vendors, tech-products, tech-categories, tech-skills, hashtags, buying-groups, board-members, years-of-experience, sub-unit-types, job-titles.

The response is keyed by field name: .<fieldName>.data[] where each entry is { attributes: { name }, id, type }. Pass the id value (not attributes.name) into search filters.

Commands

Companies

Search. Requires at least one filter.

gtm companies search --name "ZoomInfo"
gtm companies search --industry software --metro "CA - San Francisco" --employees "100to249,250to499"
gtm companies search --type public --country "United States" --revenue-min 1000000 --sort -revenue --page-size 10
gtm companies search --tech "<tech_product_id>" --metro "MA - Boston"
gtm companies search --naics "541511,541512"

Common filters: --name, --domain, --industry, --metro, --state, --country, --continent, --zip, --employees, --employees-min/-max, --revenue, --revenue-min/-max, --type, --ticker, --tech, --naics, --sic, --funding-min/-max, --funding-start/-end, --sort, --page, --page-size. Run with no flags to see the full list and a guidance error.

Enrich. Provide any identifier (most accurate: --id).

gtm companies enrich --id 344589814
gtm companies enrich --domain stripe.com
gtm companies enrich --name "Stripe"
gtm companies enrich --ticker NVDA
gtm companies enrich --domain stripe.com --fields name description website revenue employeeCount industries

# Bulk: up to 10 via --file (JSON array of identifier objects)
gtm companies enrich --file ./companies.json

The --fields flag controls which fields the MCP returns. Valid values (subset): name, description, website, phone, street, city, state, country, revenue, revenueRange, employeeCount, employeeRange, industries, primaryIndustry, companyFunding, recentFundingAmount, recentFundingDate, totalFundingAmount, foundedYear, socialMediaUrls, ultimateParentName, parentName. Omit --fields to get the defaults (name, description, city, state, country, revenue, industries).

Find similar companies (ML-ranked look-alikes — useful for territory planning, lookalike prospecting):

gtm companies similar --id 344589814
gtm companies similar --name "Stripe"   # less accurate; CLI resolves a best-match company first

Contacts

Search. Requires at least one filter.

gtm contacts search --first-name Henry --last-name Schuck
gtm contacts search --job-title "CFO OR VP Finance OR Treasurer" --employees "1000to4999,5000to9999"
gtm contacts search --management-level "C Level Exec,VP Level Exec" --company-id 344589814
gtm contacts search --management-level Manager --department "Engineering & Technical" --industry software --metro "CA - San Francisco" --required email,phone
gtm contacts search --management-level "VP Level Exec" --department Sales --accuracy-min 90

Common filters: --first-name, --last-name, --full-name, --email, --job-title (free-text OR queries OK), --exact-job-title, --management-level (canonical names only — use lookup), --department, --job-function, --company-id, --company-name, --company-domain, --industry, --metro, --state, --country, --employees, --revenue, --tech, --accuracy-min/-max (70-99), --required (subset of email,phone,directPhone,mobilePhone,personalEmail), --executives-only, --sort, --page, --page-size.

Enrich. Valid identifier combinations:

gtm contacts enrich --id 1260398587                                      # most accurate
gtm contacts enrich --email [e-mail]
gtm contacts enrich --phone [phone]
gtm contacts enrich --first-name Henry --last-name Schuck --company "ZoomInfo"
gtm contacts enrich --full-name "Henry Schuck" --company-id 344589814
gtm contacts enrich --email [e-mail] --fields phone jobTitle managementLevel email

# Bulk: up to 10 via --file
gtm contacts enrich --file ./contacts.json

Returns business-verified data only (corporate emails / business phones). The --fields flag is a comma-separated list. Default returns: firstName, lastName, email, companyId, companyName.

Similar contacts (people who look like a reference person):

gtm contacts similar --person-id 1260398587
gtm contacts similar --person-id 1260398587 --company-id 239305146   # similar contacts within Salesforce

Recommended contacts at a target company, based on your account's interaction history:

gtm contacts recommended --company-id 344589814 --use-case PROSPECTING
gtm contacts recommended --company-id 344589814 --use-case DEAL_ACCELERATION
gtm contacts recommended --company-id 344589814 --use-case RENEWAL_AND_GROWTH

Use-case picker:

• PROSPECTING (default) — cold-start friendly; based on people the user has viewed/copied/exported on the ZoomInfo platform
• DEAL_ACCELERATION — similar to contacts in closed-won opportunities (new business)
• RENEWAL_AND_GROWTH — similar to contacts in closed-won renewal opportunities

Intent signals

Every intent call requires --topics (1-50 names — use gtm lookup --field intent-topics to find exact names).

Search across all companies:

gtm intent search --topics "Cloud Applications" "Java" --signal-score-min 70
gtm intent search --topics "Mobile Apps" --signal-score-min 80 --signal-start 2026-05-01 --signal-end 2026-05-31 --industry software --country "United States"
gtm intent search --topics "AI Agents" --audience-strength-min B --audience-strength-max A

Enrich — fetch intent for a specific company:

gtm intent enrich --company-id 344589814 --topics "Cloud Applications" "Java"
gtm intent enrich --name "ZoomInfo" --topics "AI Agents" --signal-score-min 70
gtm intent enrich --website https://www.stripe.com --topics "Payments"

Signal scores range 60-100 (higher = stronger interest). Audience strength is A-E (A = largest audience researching). Use --audience-strength-min E --audience-strength-max A to include all; tighter ranges narrow to bigger groups.

Scoops (real-time business events)

Scoops capture funding, M&A, leadership moves, layoffs, product launches, partnerships, awards, hiring plans, and more. Search requires at least one filter.

gtm scoops search --scoop-types Funding --industry software --metro "MA - Boston" --published-start 2026-05-01 --published-end 2026-05-31
gtm scoops search --scoop-types "New Hire" "Executive Move" --department C-Suite --published-start 2026-05-27
gtm scoops search --scoop-types Layoffs --employees "10000plus" --published-start 2026-01-01
gtm scoops search --scoop-types "Product Launch" --description "artificial intelligence" --published-start 2026-04-01

Valid --scoop-types: Earnings, Funding, Initial Public Offering (IPO), Mergers & Acquisitions (M&A), Divestiture, Award, Event, Facilities Relocation / Expansion, Product Launch, Partnership, Hiring Plans, Open Position, New Hire, Lateral Move, Promotion, Left Company, Layoffs, Management Move, Executive Move, Pain Point, Project, Commentary, Person-Based.

Enrich — get scoops for a specific company:

gtm scoops enrich --company-id 344589814 --scoop-types "New Hire" Promotion
gtm scoops enrich --name "Stripe" --published-start 2026-01-01
gtm scoops enrich --websites https://www.zoominfo.com https://www.salesforce.com

News

Currently only enrich (no search). Requires --company-id (integer).

gtm news enrich --company-id 344589814 --categories FINANCIAL_RESULTS FUNDING
gtm news enrich --company-id 344589814 --categories PRODUCT --publishing-start 2026-03-01 --publishing-end 2026-05-31
gtm news enrich --company-id 344589814 --categories PERSON

Valid --categories: FINANCIAL_RESULTS, FUNDING, GENERAL_NEWS, GENERAL_PRESS_RELEASE, MERGER_OR_ACQUISITION, PERSON, PRODUCT.

GTM Context

The org's GTM configuration (offerings, ICPs, buyer personas, competitors, strategic priorities). Reading is free (zero credits). Updates require admin permissions.

gtm-context get                                                     # compressed (~1k tokens)
gtm-context get --detailed -f yaml                                  # full expanded profiles
gtm-context update --query "Update competitive intelligence" --source "Our main competitors are X, Y, Z…"
gtm-context update --query "Create buyer personas from these interviews" --source-file ./interview-notes.txt

The update query frames the agent's intent; --source / --source-file provides the substance to analyze (max ~2,000 words per call). When omitted, the update agent operates on existing configuration directly (good for archive, rename, simple edits).

Feedback

gtm feedback submit --category DATA_QUALITY --message "Henry Schuck's company should be ZoomInfo Technologies, not ZoomInfo"
gtm feedback submit --category FEATURE_REQUEST --message "Want a --csv-headers flag to control column order"

Categories: DATA_QUALITY, FEATURE_REQUEST, ACCESS_ENTITLEMENT_ISSUE, OTHER.

Raw — universal escape hatch

Use when you need a tool that doesn't have a curated wrapper (currently: account_research, contact_research).

gtm raw list-tools -f table                                                          # show every MCP tool the server exposes
gtm raw call account_research --args '{"companyId":"344589814"}'
gtm raw call contact_research --args '{"personId":"1260398587"}'
gtm raw call find_similar_companies --args '{"companyName":"Stripe"}'

---

Output formats

Every subcommand accepts -f, --format <format>:

Format	When to use
json (default)	Pretty-printed JSON, preserves exact response shape. Pipe to jq.
jsonl	One JSON object per line. Stream into log/data pipelines.
csv	Flat CSV with headers; one-level-nested objects are dot-flattened (company.name), arrays are JSON-stringified.
yaml	Human-readable diffs of deeply nested responses.
table	ASCII bordered table. JSON:API envelopes are auto-unwrapped and attributes is hoisted to top-level columns. Best for terminal browsing of small responses.

gtm companies search --industry software -f table
gtm contacts search --management-level "C Level Exec" -f jsonl > c-level-contacts.jsonl
gtm-context get -f yaml

For deeply nested responses, prefer json + jq or yaml. csv / table will stringify second-level-nested structures into a single cell.

---

Piping with jq

ZoomInfo responses follow JSON:API conventions: { data: [...], meta: {...} } where each item is { attributes: {...}, id, type }. So jq queries always go through .data[].attributes for the useful fields.

# Names of every result
gtm companies search --industry software --metro "CA - San Francisco" \
  | jq -r '.data[].attributes.name'

# Sorted by revenue, top 5 names + revenue
gtm companies search --type public --country "United States" --revenue-min 1000000 --page-size 50 \
  | jq '[.data[] | { name: .attributes.name, revenue: .attributes.revenue }] | sort_by(-.revenue) | .[0:5]'

# Email + phone TSV for every contact result
gtm contacts search --company-id 344589814 -f jsonl \
  | jq -r '[.attributes.email, .attributes.phone] | @tsv'

# Top 5 intent signals by score
gtm intent search --topics "Cloud Applications" --signal-score-min 80 \
  | jq '.data | sort_by(-.attributes.score) | .[0:5]'

# Lookup → extract IDs → search
INDUSTRY_IDS=$(gtm lookup --field industries --fuzzy software | jq -r '.industries.data[].id' | paste -sd, -)
gtm companies search --industry "$INDUSTRY_IDS" --metro "CA - San Francisco"

---

Common multi-step workflows

Doing many calls? Pick the cheapest option in this order:

1. Native bulk (--file) — preferred. enrich accepts up to 10 identifiers in a single MCP call, so there's no loop and no rate concern. Use it whenever you're enriching a set. (Only enrich has bulk mode — search/similar/scoops/intent/news are single-call per invocation.)

   gtm companies enrich --file ./companies.json   # up to 10 in one round trip
   gtm contacts enrich  --file ./contacts.json
   

   For >10, split into chunks of 10.
2. Bounded parallelism — alternative to the sleep loop. For commands without bulk mode (e.g. enriching from search IDs), fan out with a small concurrency cap instead of a sequential loop:

   gtm companies search --industry "$IDS" --page-size 20 \
     | jq -r '.data[].id' \
     | xargs -P4 -I{} gtm companies enrich --id {}     # max 4 in-flight
   # or: ... | parallel -j4 gtm companies enrich --id {}
   

   Keep the cap small (-P3/-P4) — unbounded fan-out trips a 429 just like a tight loop.
3. Sequential loop — fallback. If you loop one call per item, sleep at most 1 second between successive calls; never longer.

Parallelism and the 1s-sleep rule both exist to respect the subscription's per-tier rate limits, so they're mutually exclusive: use bounded parallelism or a sequential ≤1s-sleep loop, never both, and never unbounded fan-out.

Search → enrich (top results)

gtm companies search --industry software --metro "CA - San Francisco" --page-size 3 \
  | jq -r '.data[].id' \
  | xargs -I{} gtm companies enrich --id {}

Lookalikes → enrich

gtm companies similar --id 344589814 \
  | jq -r '.data[0:10][].id' \
  | xargs -I{} gtm companies enrich --id {} --fields name domain employeeCount revenue industries

Export to spreadsheet

gtm contacts search --management-level "VP Level Exec" --department Sales --company-id 344589814 -f csv > vp-sales.csv

Intent signal monitoring loop

# Daily run: companies showing strong intent for our topics, exclude those we already touched
gtm intent search --topics "Cloud Applications" "Data Warehouse" --signal-score-min 80 --signal-start "$(date -v-7d +%Y-%m-%d)" \
  | jq '.data[] | { company: .attributes.companyName, score: .attributes.score, topic: .attributes.topic }'

---

JSON Response Keys

Every MCP search/enrich endpoint returns JSON:API shape ({ data: [...], meta: {...} }). Each item under .data[] has { attributes, id, type }.

Command	Top-level key	Item shape
companies search / similar	.data[]	{ attributes: { name, city, country, revenue, employeeCount, … }, id, type:"Company" }
companies enrich	.data[] (one per input)	{ attributes: { …enriched fields per --fields }, id, type:"Company" }
contacts search / similar / recommended	.data[]	{ attributes: { firstName, lastName, jobTitle, email, phone, managementLevel, company:{id,name}, hasEmail, hasPhone, … }, id, type:"Contact" }
contacts enrich	.data[] (one per input)	{ attributes: { …enriched fields per --fields }, id, type:"Contact" }
intent search / enrich	.data[]	{ attributes: { topic, score, audienceStrength, signalDate, companyName, companyId, … }, id, type:"Intent" }
scoops search / enrich	.data[]	{ attributes: { scoopType, description, link, originalPublishedDate, companyName, … }, id, type:"Scoop" }
news enrich	.data[]	{ attributes: { title, url, publishingDate, category, snippet, … }, id, type:"NewsArticle" }
lookup	.<fieldName>.data[]	{ attributes: { name }, id, type } — pass id (not attributes.name) into search filters
gtm-context get	.result	{ user, organization, buyerPersonas:[…], idealCustomerProfiles:[…], competitors:[…], offerings:[…] }
gtm-context update	.result	varies by update operation
feedback submit	top-level	confirmation object
raw list-tools	top-level array	[{ name, description, inputSchema }, …]
raw call <tool>	varies	depends on the tool — fall back to -f yaml to inspect

Pagination lives at .meta.page.{number, total} and .meta.totalResults for searches.

---

Important behaviors

• Search commands refuse to run with zero filters. Running e.g. gtm companies search with no flags exits with a guidance error listing common filters. This is intentional — unfiltered searches return unbounded uninteresting results.
• Filter values must be canonical IDs / names from the lookup endpoint. Free-text industry / metro / topic strings return HTTP 422.
• Bulk enrich caps at 10 entries per call. Prefer enrich --file (one MCP call for up to 10 identifiers) over looping. For larger sets, batch into chunks of 10.
• For high-volume calls, prefer in this order: (1) native bulk enrich --file; (2) bounded shell parallelism (xargs -P4 / parallel -j4); (3) a sequential loop sleeping at most 1 second between calls. Use bounded parallelism or the ≤1s-sleep loop, never both, and never unbounded fan-out — both guards exist to respect the subscription's rate limits. See the callout at the top of Common multi-step workflows.
• Some MCP tools double-encode their response payload (the text content block is a JSON string of a JSON object). The CLI handles this transparently — agents shouldn't need to care.
• lookup for tech-products requires a two-step flow: first lookup tech-vendors with a fuzzy match, then lookup tech-products with --vendor "<exact vendor name>". Passing --field tech-products without --vendor returns nothing useful.
• Compliance: This skill drives ZoomInfo data. Apply data-minimization (only fetch what you need), respect the rate limits of the user's subscription tier, and never use the data for unsolicited bulk outreach without a legal basis. ZoomInfo's Terms of Service apply to everything accessed through this CLI.