Jorge LewisA technical founder can run a first LinkedIn outbound campaign from a coding agent: add the GTM Bud MCP server with one command, describe one buyer in a sentence, and the agent builds the campaign. Launch waits for your confirmation of a preview, and the agent shows you every reply before it sends. You should not do this yet if you cannot name the buyer or have no time to answer replies. Both cases are covered below.
This guide is for the technical founder of an early B2B SaaS company who has never done outbound and would rather stay in a terminal than learn a sales tool. It uses GTM Bud through its connector. If you are a GTM engineer who clones campaigns for many segments, the page on AI LinkedIn outreach for GTM engineers fits you better.
Who should not do this yet?
Do not start a first outbound campaign if one of the two conditions below is true. No tool fixes either one, and a campaign makes both worse.
| Condition | Why it blocks you | What to do first |
|---|---|---|
| You have no clear buyer | The campaign starts from one audience sentence. If you write “engineering teams”, nothing in it is a person, and the prospect preview will show you the wrong people. No check can choose the buyer for you | Talk to the five users who got the most value and find the job title they share |
| You have no time to answer replies | The agent shows you every reply for review before it sends. An Interested lead who waits a week is a lost conversation | Wait until you can give replies a fixed slot every working day |
A third case is softer. If nobody can use the product without you on a call, outbound will fill your calendar with onboarding. That can be fine at five customers. Decide it on purpose.
If you pass both tests, the rest of this guide is the campaign.
What does a first campaign from a coding agent look like?
A first campaign from a coding agent is a single session in which the agent calls GTM Bud tools and you make five decisions. The agent does not replace your judgment. It removes the forms. You decide the audience sentence, which previewed prospects fit, which value offer you can really deliver, whether the sequence says anything false about your product, and whether to confirm the launch preview. Between those decisions the agent creates the draft, reads the criteria check, starts the prospect preview, fetches the GTM Bud instructions for value offers and for the message sequence, writes both, and asks GTM Bud to check and save them. Then it runs the launch readiness check, which returns a list of what is missing. Nothing in this flow sends a message until you confirm a preview. After launch, the same session is where you read results and review replies.
Step 1: Add the server and write one sentence
Setup is one command.
claude mcp add --transport http gtm-bud https://gtmbud.com/mcpSign in to GTM Bud in the browser window that opens. Steps for Cursor and Codex are on GTM Bud for coding agents.
Then write the audience sentence. Engineers tend to write a query. Write a description.
| What a technical founder writes first | The problem | A sentence that works |
|---|---|---|
| “Teams running dbt in production” | A team is not a person, and “in production” is not on a profile | “Heads of data and data platform leads at B2B software companies with 50 to 300 employees in the US” |
| “Anyone with slow CI” | No role, no company | “Engineering managers at mobile app companies with 30 to 200 employees in Europe” |
| “CTOs” | Every CTO on earth | “CTOs at logistics software companies with 20 to 100 employees in the US” |
Two filters do not belong in the sentence. GTM Bud sourcing already screens for recent LinkedIn activity and already skips people you have contacted, so those are not criteria.
Step 2: Read the criteria check like a failing test
Treat the criteria check as a test report on your sentence. create_campaign_draft returns the draft and the check. Your agent turns the sentence into criteria. GTM Bud checks each one and flags the ones that cannot be read from a profile, a company page or public web content, with a suggested rewrite.
Technical founders usually get a flag on a tool qualifier. You know that your buyer uses a certain tool. A LinkedIn profile often does not say so. When the check flags that qualifier, you have two honest choices:
- Drop it from the targeting and judge it by eye in the preview.
- Reword it to something a profile shows, such as a job title that implies the tool.
Then call the preview. start_prospect_preview and get_prospect_preview return real matching people. You say which ones fit, and the agent records it with mark_prospect_fit. If you say Not a fit more often than Good fit, fix the sentence before you write anything. The longer method is in how to build an ICP for outbound that converts.
Step 3: Offer something you can ship this week
A value offer is what the prospect gets for answering. For an early SaaS product, the honest offer is a small output of the product itself on the prospect’s own data.
The agent calls get_value_offer_brief, researches your site, writes offers, and save_value_offers has GTM Bud check and save them. Your review has two questions:
- Is every technical claim true today? Not on the roadmap. Today.
- Can you deliver it alone, ten times, this month? If it needs a custom integration, it is a project, not an offer.
The message sequence follows the same pattern with get_sequence_brief and save_campaign_sequence. Read it line by line. A check can catch a broken format. Only you can catch “works with every warehouse” when it works with two.
Step 4: Launch, then give replies a daily slot
Launch happens in two steps: the agent shows a launch preview with the first message of each channel, and GTM Bud launches only after you confirm. Before that, get_launch_readiness lists what is missing. Expect it to name the things that live on gtmbud.com: a connected LinkedIn account, a sending account for the campaign, and the sending schedule. The 14-day free trial starts when you launch your first campaign. GTM Bud focuses on Sales Navigator accounts.
After launch, replies are the work. The agent reads them with list_replies and get_conversation, labels each one Interested, Warm or Not interested, and writes an answer. review_reply shows you the answer. GTM Bud instructs the agent to call send_reply only after you approve the exact text. The limit is 40 replies a day for each LinkedIn account, which a first campaign is unlikely to reach.
Put the review in your day like a standup. Ten minutes, the same time, every working day.
A session transcript
This is a realistic first session in Claude Code. The tool names are real. The data is invented, and results are left out because they would be your numbers.
you Draft a campaign for heads of data and data platform leads at
B2B software companies with 50 to 300 employees in the US
that run dbt.
agent create_campaign_draft
Draft created. Criteria check: roles, size and region are
clear. "Run dbt" cannot be read from every profile.
you Drop it from targeting. I will check it in the preview.
agent update_campaign_targeting
Targeting saved.
you Set the seller profile of this draft from our site and my
LinkedIn profile.
agent update_seller_profile
Saved: the website and the LinkedIn profile URL.
agent start_prospect_preview
agent get_prospect_preview
Real matching prospects returned for review.
you First one is a fit. Second one is not, that is a BI analyst.
agent mark_prospect_fit
1 marked Good fit, 1 marked Not a fit.
agent get_value_offer_brief
agent save_value_offers
GTM Bud checked and saved 2 value offers.
agent get_sequence_brief
agent save_campaign_sequence
GTM Bud checked and saved the sequence.
you What blocks launch?
agent get_launch_readiness
Missing: no LinkedIn account is connected. Connect it on
gtmbud.com, then choose it as the sending account.A worked example: a data tooling startup
The company in this example has two founders and sells a tool that finds slow and failing models in dbt projects. It has nine paying customers, all from the founders’ network. The names are invented.
The audience sentence. “Heads of data and data platform leads at B2B software companies with 50 to 300 employees in the US that run dbt.”
What the criteria check said. Roles, size and region were clear. “Run dbt” cannot be read from every profile. The check flagged it as hard to verify from public data. The founder dropped it from the targeting and checked it by eye in the preview.
Two previewed prospects.
- Nadia F., Data Platform Lead at a logistics software company in Denver with about 120 employees. Her profile lists dbt and a warehouse migration last year. Marked Good fit.
- Kevin O., Senior BI Analyst at a marketing software company in Boston with 210 employees. Marked Not a fit: he builds dashboards and does not own the pipeline, so he cannot install anything.
One value offer. “A report of the ten slowest models in your dbt project, with the cause for each, from one read-only run.” The founder can produce it in an hour with the product as it is today. He rejected a second offer that promised cost savings in dollars, because the tool does not measure cost yet.
One reply draft. A lead wrote: “We already get run times from our orchestrator. What does this add?” The agent wrote, for review:
“The orchestrator tells you which model is slow. The report tells you why: for example a full refresh that could be incremental, or a join that fans out. It needs read access for one run and nothing is written to your project. Want me to send the two setup lines?”
The founder approved it, and the agent sent the approved text.
The results question. “Get the results for the data platform campaign as of now. Then list every reply labeled Not interested and group them by the reason they gave.”
Five prompts for a technical founder
- “Create a campaign draft for heads of data and data platform leads at B2B software companies with 50 to 300 employees in the US. Print the criteria check in full.”
- “Set the seller profile of this draft from
README.mdand our landing page. Show me the website and profile URL before you save.” - “Start a prospect preview. For each person, show role, company and size, and ask me fit or not fit one at a time.”
- “Get the value offer instructions and write two offers. Use only features that exist in the
mainbranch today.” - “Every morning when I ask for replies: label the new ones, show me drafts for the Interested ones, and send nothing until I say yes to each.”
Three objections technical founders raise
“I do not want to be a spammer.” Then be strict on two things that you control: the audience and the offer. Mark every doubtful prospect Not a fit, and offer something a data lead would pay for. The limit is that first messages and follow-ups go out from the sequence you approved before launch, not one by one. If that sequence reads like a pitch, rewrite it before you confirm.
“Why can I not automate the replies too? That is the boring part.” Because a reply is where a real person asked a real question about your product, and a wrong answer under your name costs more than a slow one. The agent writes each reply. You approve each one. That is by design, and it is the reason for the second test at the top of this page.
“This is not really all in the terminal.” Correct. Connecting a LinkedIn account, choosing the sending account and setting the schedule are done on gtmbud.com. You do them once per account or campaign. The readiness check tells you in the terminal which one is missing.
Frequently asked questions about a technical founder’s first outbound
Can I run LinkedIn outbound from Claude Code?
Yes. Add the GTM Bud MCP server with claude mcp add --transport http gtm-bud https://gtmbud.com/mcp and sign in to GTM Bud in the browser. The agent can then create a campaign draft, preview prospects, write offers and a sequence, and launch after you confirm a preview.
Should a technical founder do outbound before product-market fit?
Yes, if you can name one buyer and have time to answer replies. Outbound is a fast way to get conversations with that buyer. If you cannot name the buyer yet, do customer interviews first, because a campaign needs one clear audience to start. For the sales habits around those conversations, read founder-led sales.
Can I script the whole thing and run it on a schedule?
No. The connector is built for a session with you in it. Launch needs your confirmation of a preview, each change that starts sending needs your confirmation, and the agent shows each reply for your review before it sends.
Do I need to write the outreach messages myself?
No. GTM Bud gives the agent its instructions for the message sequence, the agent writes it, and GTM Bud checks and saves it. Read the sequence before launch. You know the product best and you will see a wrong technical claim faster than any check.
What can I not do from the terminal?
Three things are done on gtmbud.com: connecting a LinkedIn account, choosing the sending account for a campaign, and setting the sending schedule. The 14-day free trial starts when you launch your first campaign. The full list of what the connector can do is on the GTM Bud connector page.
Ship one campaign, then read what comes back
For a technical founder, first LinkedIn outbound is closer to a small release than to a sales project: one command, one sentence, five decisions, and a daily review. Do it when you can name the buyer and answer the replies, and wait when you cannot. Start with the setup on GTM Bud for coding agents. When the first campaign is running, outbound sales for startups without an SDR covers the next month, and lead generation for SaaS companies shows where LinkedIn fits with your other channels.