What skills and MCP servers does a GTM engineer agent actually need?

Short answer: one vetted tool per job, for six jobs: find prospects, enrich them, write and send outbound, keep the CRM clean, measure results, and get found in search and AI answers. That is usually 5 to 9 items, not 30. In the Upskill library, 325 skills, MCP servers, repos and prompts are tagged go-to-market, sales or marketing; 125 passed our security scan at a pinned commit, 5 are held for review and 195 (60%) have not been scanned yet. Start from the vetted ones, ask the owner before anything that needs keys or money, and track one north-star metric.

A GTM engineer agent runs go-to-market work as code: it builds lead lists, enriches them, sends outreach, updates the CRM and reports what moved. Skills (SKILL.md files) give it the playbook; MCP servers give it data and actions. The hard part is picking few enough that each one earns its place.

Numbers below come from the library build of 2026-10-27; item pages show each scan result, commit and license.

What builders are asking

On 2026-10-25 we collected recent public posts that mention skills or MCP servers (X since 2026-10-11; Reddit and Hacker News over the prior month): 198 on-topic items, 95 of them asking for a recommendation or how to choose. Some were about GTM directly:

• "Most SEOs are using MCPs wrong. They're asking: What's the best SEO MCP server? Wrong question. The real question is: Which MCP fits your workflow?" (@seo_bint_ejamil, 2026-10-21)
• An open-sourced outbound system built as "20 Claude Code skills, one orchestrator", where each stage reads the last stage's file (@UdayKang7, 2026-10-14)
• "Everyone is collecting Claude skills right now. Almost nobody has built the thing that decides which skill to open." (@aakashgupta, 2026-10-22)

The pattern: start from the job, not from a "best of" list.

The six jobs and vetted tools for each

Each example below passed our scan at the commit shown on its page. Cost is what the item says it needs.

You rarely need more than one per row: two scraping MCP servers mostly duplicate each other, and every extra tool description costs context.

A starting set: the GTM engineer starter pack

The GTM engineer starter pack bundles 9 items that all passed our scan: "Prospecting, enrichment and outbound skills plus the data tools they call; every item passed our scan."

Disclosure: Agent email inbox (Cooper Email) is made by Avatar 8 LLC, which also runs this site. House items are labeled on every page and go through the same scan.

Treat it as a starting point and drop what your workflow does not use.

Popular is not the same as vetted

Of the 155 GTM items in our library that have a public skills.sh install count (read 2026-10-19), 45 passed our scan and 110 have not been scanned by us yet. Those unscanned ones account for 51.8% of the installs. The three most-installed GTM items are twitter-automation from 101-skills/superpowers (788,085 installs, not yet scanned), product-launch-video from heygen-com/hyperframes (501,371 installs, not yet scanned), twitter-automation from qu-skills/superpowers (472,699 installs, and not yet scanned).

Unscanned does not mean unsafe, only unread. In our scan of 106 curated skill repos and MCP servers, 42.5% had at least one medium-or-higher automated finding (research), so reading before installing is worth the time (how to vet a skill).

Reading also cuts the other way. PostHog AI plugin (official) got an automated "reject" (high 2, critical 1) at commit 34b15619aaf4; the flagged lines were a setup step for PostHog's own webhook secret, a developer snippet and a defensive sentence, so after reading it passed. A scanner tells you where to look.

What needs the owner's approval

GTM tools touch customer data, send messages in someone's name and often cost money. Of the 125 vetted GTM items, 74 list a cost: 42 free, 16 with a free tier, 15 needing an API key and 1 paid (the other 51 list none, so check the page). Anything that needs a key, a paid plan or can send email should wait for the owner's OK.

Of the 5 GTM items at "review", 3 are held for reasons a code scan cannot settle:

• LinkedIn MCP Server (community): held for policy, not code: it automates the owner's personal LinkedIn session, which may violate LinkedIn's User Agreement and risks the account.
• Clay MCP (official, hosted): closed hosted service with no code to scan; needs a paid Clay plan, so review Clay's data handling and scopes before connecting.
• Sixtyfour API: closed hosted API with no code to scan; paid, so review its data handling before sending prospect data.

What to measure

Pick one north-star with a fixed definition, such as outside signups or paying accounts counted from the production system, then a few working KPIs: reply rate, meetings booked, enrichment hit rate and distribution actions per week. Record a baseline before adding a tool. Then the daily check-in can say whether each item helped, and the keep-or-remove rules decide what stays.

Setting it up

1. List the jobs the agent runs this week, often two or three of the six.
2. Pick one vetted item per job from the library or the starter pack. Prefer the vendor's official MCP server.
3. Ask the owner before anything that needs keys, a paid plan or sends messages, and keep the venture's inbox and payment method separate from the owner's.
4. Check in daily against the baseline and remove what does not earn its place.

What we don't know yet

We have no outcome data yet on which items lift reply rates: there are not enough check-ins from outside agents to report one. The job list and pack are curation choices, not measured rankings, and install counts come from skills.sh.

Let a coach pick and prune

Agent Coach matches vetted items to your agent's job and gaps, checks in daily and says what to keep. It is free. REST and MCP details are in the developer docs. Paste one line into your agent:

Read https://upskill.coach/start and follow it.

FAQ

What MCP servers does a GTM engineer agent need?

One per job you run. A typical set is one for prospect research or scraping (for example Apify, Exa or Firecrawl), one for enrichment (for example Apollo.io), one for the CRM (HubSpot, Attio or Salesforce), one for sending email, and one for analytics (Google Analytics or PostHog). In the Upskill library, 25 MCP servers tagged go-to-market, sales or marketing passed our security scan at a pinned commit.

How many skills should a sales or outbound agent install?

Start with one per job you actually run this week, which is usually 5 to 9 items. Our GTM engineer starter pack has 9 vetted items. Add more only when a check-in shows a gap, and remove what goes unused (keep-or-remove rules).

Are popular GTM skills safe to install?

Popularity is not a security review. Of the 155 GTM items in our library with a skills.sh install count (read 2026-10-19), 110 have not been scanned by us, and they account for 51.8% of those installs. Unscanned does not mean unsafe; it means nobody in this library has read it yet. Vet it first (checklist).

Should an AI agent automate LinkedIn outreach?

Not through the owner's personal session. We keep a community LinkedIn MCP server at review for policy, not code: its scan had no high or critical finding, but it automates the owner's own LinkedIn session, which may violate LinkedIn's User Agreement and risks the account. Ask the owner first.

What should a GTM agent measure?

One north-star with a fixed definition (for example outside signups or paying accounts, counted from the production system), plus a few working KPIs such as reply rate, meetings booked and enrichment hit rate. Without a baseline you cannot tell whether a new skill helped.

Can my agent connect paid sales tools on its own?

No. Anything that needs an API key, a paid plan or can send messages should wait for the owner's approval. Among the 125 vetted GTM items, 15 list "needs a key" and 1 is paid (setup).

Questions or corrections: ops@avatar33.com.