Jorge LewisMCP for sales is worth adopting only when the connector passes six checks: its tools are marked read or write, you know what stands between the AI and a send, limits are enforced by the platform, the writing instructions come from the platform, results come with definitions, and you can stop a campaign in one request. This guide gives RevOps leads and GTM engineers the questions to ask of any sales connector, with the GTM Bud connector as the worked example.
It is written for the person who has to say yes or no to a connector, not for the person who wants to know what MCP is. That part takes one paragraph.
What is MCP, in one paragraph?
MCP, the Model Context Protocol, is an open standard that lets an AI app use tools and data from another product through one common interface. Anthropic published it in late 2024, and it is documented at modelcontextprotocol.io. A product publishes an MCP server once, and any AI app that supports MCP can connect to it, usually with a URL and a sign-in. For sales, the effect is that an assistant no longer sees only what you paste into the chat. It can ask your outbound platform for replies, change a campaign and ask for a launch. That is why the evaluation matters.
The six questions at a glance
| Question | A good answer | A bad answer |
|---|---|---|
| 1. Which tools read and which write? | Each tool is marked read or write | One broad tool that does everything |
| 2. What stands between the AI and a send? | A preview and a second call, or a stated review step | The AI calls a tool and messages go out |
| 3. Who enforces the limits? | The platform, with a stated number | A sentence in the prompt |
| 4. Where do the writing instructions come from? | The platform gives a brief and checks the result | Whatever the AI decides in the session |
| 5. What do the analytics return? | Counts, rates with definitions, and a time | A percentage with no source |
| 6. How do you stop? | One request, applied at once | A support ticket or a different app |
1. Which tools read, and which tools write?
Ask for the tool list and look for a read or write mark on each tool. A connector with clear marks lets you give a team a simple rule: read tools are free to use, write tools follow the process.
In the GTM Bud connector, each tool is marked read or write. Three groups make the list easy to review. These are examples from each group, not the full list:
- Read.
list_campaigns,get_campaign_results,list_replies,get_conversation,get_reply_playbooks,get_launch_readiness,get_campaign_controls. These change nothing. - Write to a draft.
create_campaign_draft,update_campaign_targeting,duplicate_campaign,save_value_offers,save_campaign_sequence. A draft sends no messages. - Write to something live.
send_reply,launch_campaign,set_sending_mode,set_lead_finding,end_campaign.
A GTM engineer can check this in two minutes: connect the server in a coding agent and ask it to list the tools with their marks. If the list does not match the documentation, stop the evaluation there.
2. What stands between the AI and a send?
Ask what stands between a tool call and a message to a real person, and ask which kind of control it is. A preview with a second call is a control in the platform. An instruction to the AI is a request. A vendor should tell you which of the two protects each action.
With GTM Bud the answer has two parts:
- Changes that start something. Launch, a change that starts sending, a change that starts lead finding and ending a campaign each need two calls. The first call returns a preview. The change applies only after the user confirms and the AI makes the second call.
- Replies. The AI writes the text and calls
review_reply, which shows it in the chat for you to edit and send.send_replyis a separate write tool that GTM Bud instructs the AI to call only after you approve the exact text. Unlike launch, it has no preview and second call, so your rule for who approves replies is the control.
That is the kind of answer to look for from any sales connector: which actions the platform holds behind a second call, and which actions rest on an instruction to the AI and on your own process. An answer that claims a lock on everything deserves a look at the tool list.
Ask the vendor one more question: who is “the user”? In GTM Bud it is the person signed in to the account in that AI session. There is no approval chain with roles, so a rule such as “the author does not confirm their own launch” is a rule in your process.
3. Who enforces the limits?
The platform must enforce the limits, and it must state the number. An AI that promises to “send slowly” has no way to keep the promise across sessions.
GTM Bud limits replies to 40 a day for each LinkedIn account. One GTM Bud account can hold several LinkedIn sender accounts, and a campaign can be assigned specific senders, so the limit scales with the accounts and not with the number of AI sessions. The sending schedule is set on gtmbud.com and not in chat, which keeps it out of reach of a careless prompt.
4. Where do the writing instructions come from?
Ask where the AI gets its instructions for a message, a value offer or a reply. If the answer is “from the model”, every session writes to a different standard and RevOps has nothing to inspect.
GTM Bud uses the same pattern for value offers and for the message sequence:
- The AI asks GTM Bud for its instructions (
get_value_offer_brief,get_sequence_brief). - The AI researches and writes.
- GTM Bud checks the result and saves it (
save_value_offers,save_campaign_sequence).
Replies follow reply playbooks: one for the account and one per campaign. The AI reads them with get_reply_playbooks. For a change, it shows the exact new text and saves it only after the user approves it in the chat, and each save keeps the earlier text in the version history. For a RevOps lead this is the important part. The standard is text that you can read and change, and it is the same text in every session.
5. What do the analytics return?
A rate without its parts is not analytics. Ask if each rate comes with the numbers behind it, a definition, and the time it was calculated.
get_campaign_results returns results for one campaign: each rate with its numerator, its denominator and its definition, and an as-of time. The honest limits are part of the answer. Results are per campaign, so a view across campaigns is one call for each campaign. There is no scheduled report. The numbers refresh when someone asks and can be up to about 15 minutes old; the as-of time gives the exact time. Our guide to LinkedIn outbound results in your terminal shows how to turn that into a weekly review.
6. How do you stop?
You should be able to stop sending in one request, without a preview, from the same place you started. set_sending_mode has three settings: all, follow-ups only and stopped. A change that stops sending applies at once. Ending a campaign is a different kind of change, and it asks for a confirmation of a preview.
A worked example: one evaluation session
The company in this example sells identity management software to universities. The RevOps lead and a GTM engineer run the evaluation together, and nothing is sent. The names below are invented.
The audience sentence. “Heads of IT at private universities with 2,000 to 15,000 students in the US Northeast.”
What the criteria check said. The check passed the role and the region. It flagged student numbers as hard to verify from public data and suggested a rewrite. The AI then asked which titles “head of IT” covers, and proposed CIO, VP of IT and Director of IT. It also listed the states it would treat as the Northeast and asked for confirmation.
Two previewed prospects.
- Robert K., CIO at a private university with about 6,000 students in Massachusetts. Marked Good fit: right role, right kind of institution, right size.
- Hannah L., IT Support Manager at a public community college in New Jersey. Marked Not a fit: a public college, and a role that does not own the identity system.
One value offer. “A review of how student and staff accounts are created and removed at your university, with a written list of accounts that stay active after a person leaves.” The AI asked GTM Bud for its instructions, researched the company website, wrote the offer, and GTM Bud checked and saved it.
The results question. There were no results, because nothing was launched. The engineer asked the question on an older campaign: “Give me the results for the campaign that ended in the spring, with the definition of each rate and the as-of time.” The answer is what they compared with the dashboard.
What the session proved. All six questions had an answer that the two of them could see in the session, and the draft sent no messages.
Five prompts for an evaluation session
- “List every tool the GTM Bud server gives you. For each one, say if it is marked read or write, and group the write tools by what they can start.”
- “Create a campaign draft for heads of IT at private universities with 2,000 to 15,000 students in the US Northeast. Show me the criteria check word for word.”
- “Ask GTM Bud for the value offer instructions and show them to me before you write anything.”
- “Run the launch readiness check on the draft and tell me which missing items I must fix on gtmbud.com.”
- “Show me the controls for a launched campaign and explain which changes would need my confirmation. Do not apply any.”
Three objections from the people who evaluate
“A connector is one more surface to secure.” True. The connector uses your GTM Bud sign-in and acts only on the account that you allowed. Your AI app receives the data that a tool call requests, and it handles that data under its own terms. Review both sides, and remove the connector in the AI app settings when a person leaves the team.
“We need this to run without a person.” Then it is the wrong tool for that requirement. The connector is built for a session with a person in it. There are no webhooks and no API beyond the connector on offer here, and no list upload through the connector. A scheduled job that launches campaigns is not possible, by design.
“Not everything is in the chat.” Correct. Connecting LinkedIn accounts, choosing sending accounts and the sending schedule are done on gtmbud.com. Count that as part of the setup time. The 14-day free trial starts when you launch your first campaign.
Frequently asked questions about MCP for sales
Is an MCP connector the same as a Zapier style integration?
No. A Zapier style integration runs a fixed trigger and action. An MCP connector gives an AI a set of tools it can choose from during a conversation, so the AI decides which tool answers a request. That choice is the reason to check which tools can write.
Does an MCP connector give the AI access to everything in my account?
It gives access only to the tools the connector exposes, and only after you sign in and allow it. Read the tool list after you connect, and remove the connector from your AI settings if you stop using it. The GTM Bud connector page lists what the tools cover.
Who should own sales connectors in a company?
RevOps should own the list of approved connectors, the rule for who approves a reply and who confirms a launch, and the reply playbooks. A GTM engineer can own the technical review of the tool list. Reps then use the connectors inside those rules. The RevOps checklist for adding AI to outbound covers those decisions.
Can I evaluate a sales MCP connector without sending anything?
Yes. Connect it, list the tools, and use only the read tools and the tools that create drafts. With GTM Bud a draft campaign sends no messages, so you can test targeting, previews and the readiness check with no contact to real people.
Does a sales MCP connector replace an API or webhooks?
No. A connector is an interface for an AI in a session with a person. It is not a data feed. If your requirement is a system-to-system integration that runs with nobody present, a connector is the wrong tool for that requirement.
Vet MCP for sales with six questions, then connect
MCP for sales pays off when the connector shows you what can write, tells you exactly what stands between the AI and a send, enforces its own limits, gives the AI its instructions, returns numbers with definitions, and stops in one request. Run the evaluation session above with drafts only. Setup is one sentence: add https://gtmbud.com/mcp with the steps on GTM Bud for Claude, GTM Bud for ChatGPT or GTM Bud for coding agents. Then read the page for your role: AI LinkedIn outreach for RevOps or AI LinkedIn outreach for GTM engineers. If you are comparing whole platforms and not only connectors, see how to choose outbound sales software. New to GTM Bud? You can create an account in the sign-in window.