By Shankar Ganapathy, Co-founder, Boomerang AI. Published November 16, 2026.
For twenty years, buying a sales tool meant buying a screen. A login, a tab, a seat licence, and a rep who had to remember to open it.
That is changing fast. The dialer, enrichment, signals, relationships, sequencing and orchestration are all turning into capabilities that other software calls.
The teams moving fastest are not buying more apps. They are building one surface around how they sell, and plugging the best capability into each step.
This is not about teams being lazy or cheap with "vibe coding." It is the opposite. It is teams finally making the stack fit the process, instead of bending the process to fit ten different tools.
What we are seeing in the market
The vendors are going headless
In the last twelve months, most of the systems a sales team relies on became callable by an AI assistant or agent through MCP (Model Context Protocol), not just through a human in a browser.
| Vendor | What shipped | When |
|---|---|---|
| Salesforce | Hosted MCP servers, generally available (Enterprise Edition and above) | April 2026 |
| HubSpot | Remote MCP server, generally available to all accounts, with write access to CRM records | April 2026 |
| Apollo | Native connector for Claude (beta), then an app in ChatGPT | February and April 2026 |
| 6sense | MCP server for its ABM platform | July 2026 |
| Unify | Relaunched as what it calls "Claude for outbound sellers" | June 2026 |
| Databook | Repositioned as a "headless, composable system" with API and MCP | 2026 |
Salesforce's own developer program uses the word directly: its MCP work sits under a project it calls Headless 360.
When the system of record, the contact database, the intent platform and the outbound engine all expose themselves as APIs, the screen stops being the product. The capability is.
The buyers are building
The other half of the shift is on the buying side.
- 35% of teams have already replaced at least one SaaS tool with a custom build, and 78% expect to build more internal tools in 2026, according to Retool's Build vs Buy report (817 builders, February 2026). CRM tools were among the categories most often replaced, at 25%.
- 65% of senior revenue operators have postponed a software purchase because they believed they could build it faster with AI, in a Fullcast survey of nearly 100 leaders, 97% of them Director level or above.
- 71% of GTM engineers use AI coding tools, alongside Clay at 84% and their CRM at 92%, in the State of GTM Engineering 2026 survey of 228 practitioners.
- In ColdIQ's 2026 GTM tool report of 62 revenue leaders, Lovable, a tool for building software, made the top ten at 29%.
These are mostly vendor surveys with modest samples, so read them as direction, not precision. But the direction is consistent.
Why this is happening now
AI made building cheap. A RevOps lead who could not get engineering time can now build a working internal app in days with Claude Code, Cursor or Lovable.
Agents need APIs, not screens. An AI assistant cannot click through ten tabs. It can call ten tools.
Once your team does real work in Claude, ChatGPT or an agent, every tool that is only a screen becomes friction.
Every company sells differently. The territory model, the account tiers, the hand-offs, the words the team uses.
Off-the-shelf tools make every team sell like the average customer. A custom surface lets you sell like yourself.
What a headless sales stack looks like
Picture one workspace your team actually opens, built around your own process. Behind it, each job is a capability you call:
| Job | Called as a capability |
|---|---|
| System of record | CRM over API or MCP |
| Who to target | Signals and intent |
| Who they are | Enrichment and contact data |
| How to get in | Relationships and warm paths |
| What to say | Your own AI, with your context |
| How to reach them | Sequencing, dialer, email |
| Did it work | Your warehouse and reporting |
We see this in our own customers. Digideck, a Boomerang AI customer, connected Boomerang AI to Claude through the Boomerang MCP and had Claude map which of its 6,300+ customer and prospect relationships knew which buyer, inside a workflow its own team designed (case study).
The relationship data came from us. The surface was theirs.
CLIs are not the future either. They are glorified APIs.
Right now the loudest debate among builders is CLI versus MCP. AI coding agents work in the terminal, so vendors are rushing to ship command line tools, and plenty of engineers argue a good CLI beats an MCP server for agents.
Our view is less popular: the CLI is a phase, not a destination.
A CLI is a thin wrapper around an API. Take away the API and the CLI does nothing. Take away the CLI and the API still works.
The same is true of every surface before it: desktop apps, web apps, mobile apps, Slack bots, chat, and now MCP servers and terminals. Each one was the future for a while. The API underneath each of them stayed.
So when a vendor tells you their CLI or their MCP server is the strategy, ask what it wraps. If the answer is a thin API over thin data, the new door does not change much. If the answer is data and judgment you cannot get elsewhere, the door barely matters, because your team will pick whichever door its agents use this year.
UI, CLI, MCP, chat. Those are doors. The API is the building.
That is where the market is moving, and it is what you are really paying for.
What survives, and what does not
Not everything gets rebuilt. Our read:
What survives
- Data nobody else has. Proprietary data, and relationship data that takes years to collect and verify, is not something a weekend project recreates.
- Hard judgment. Scoring, ranking and deliverability, where being right depends on scale and feedback loops.
- The system of record. Teams will build around their CRM, not replace it, in most cases.
What does not
- The screen you log into because the contract says so. If a tool's main value is its interface, and an assistant can do the same job by calling its data, the interface is the part at risk.
- Seat-based pricing for software an agent uses. When an agent does the work, seats stop matching value. Usage and outcome pricing is already spreading, from Rox's metered Agent Actions to Databook's outcome-based model.
The fair counterpoint
Building is cheap. Maintaining is not.
In the same Fullcast survey, leaders who lean buy still outnumber those who lean build by about three to one, and about half say the answer depends on the use case. Retool's report also found that 60% of builders shipped tools outside IT oversight in the past year, which is a governance problem waiting to happen.
And fully autonomous selling is not here yet. In the 2026 State of B2B GTM report, agentic campaigns ranked last of 16 GTM channels for adoption, at 7% (report).
So this is not "build everything." It is a narrower, more useful idea.
Buy the capability. Own the surface.
What this means for revenue leaders in 2027 planning:
- Ask every vendor for the API and MCP story first. If a tool only works through its own screen, it will not work with your agents.
- Pay for data and judgment, not interfaces. The question for each renewal is what you would lose if you only had the API.
- Give one person ownership of the surface. A GTM engineer or RevOps lead who designs how the pieces fit, and maintains it.
- Keep the system of record boring. Build around your CRM, not instead of it.
- Govern what gets built. Decide who can ship tools that touch customer data, and how they are reviewed.
The sales stack is not getting smaller. It is getting quieter. Fewer screens, more capabilities, and one surface that looks like your business.
Where Boomerang AI fits
Relationships are one of the capabilities that are hard to rebuild: who on your team, among investors and board, customers and partners, knows someone at a target account, and how well. Boomerang AI exposes that as an MCP server and an API, so warm paths, the recommended sender and coverage gaps can live inside whatever surface your team builds, from Claude to Clay to your own app. MCP and API access come with the Team, Company and Enterprise plans (MCP and API).
Read next: What is a GTM context graph? and Signals tell you who. Intros get you in.
Sources
- Salesforce Developers, Salesforce hosted MCP servers are now generally available, April 2026: https://developer.salesforce.com/blogs/2026/04/salesforce-hosted-mcp-servers-are-now-generally-available
- Salesforce Developers, Headless 360 MCP server beta, July 2025: https://developer.salesforce.com/blogs/2025/07/announcing-the-headless-360-mcp-server-beta
- HubSpot Developers, remote HubSpot MCP server is now generally available: https://developers.hubspot.com/changelog/remote-hubspot-mcp-server-is-now-generally-available
- Apollo MCP: https://www.apollo.io/product/mcp ; Apollo connector for Claude, PR Newswire via TMCnet, 24 Feb 2026: https://www.tmcnet.com/usubmit/-apolloio-delivers-gtm-outbound-execution-claude-/2026/02/24/10337366.htm
- 6sense launches MCP server, Business Wire, 14 Jul 2026: https://www.businesswire.com/news/home/20260714566309/en/
- Unify, announcing Unify's next chapter, June 2026: https://www.unifygtm.com/blog/announcing-unifys-next-chapter
- Databook, The GTM Decision System: https://databook.com/
- Retool, 2026 Build vs Buy report: https://retool.com/blog/ai-build-vs-buy-report-2026
- Fullcast, Builder, Buyer or Both: https://www.fullcast.com/ebooks/builder-buyer-both-ebook/
- State of GTM Engineering 2026: https://stateofgtme.com/
- ColdIQ, The 2026 GTM Tool Report: https://coldiq.com/blog/the-2026-gtm-tool-report-what-62-revenue-leaders-run
- Rox pricing (Agent Actions): https://www.rox.com/pricing
- 2026 State of B2B GTM report, Kyle Poyar and Maja Voje, p.16: https://stateofb2bgtm.com/
- Digideck case study: https://www.getboomerang.ai/customers/digideck-claude-boomerang-mcp-abm
Survey figures are from the named reports, several of them vendor surveys with modest samples. The predictions about what survives are our view.
Frequently asked questions
What is a headless sales stack?
A sales stack where each tool, such as CRM, enrichment, signals, relationships and sequencing, is used through an API or MCP server rather than its own screen. The team works in one surface it designs, often an AI assistant or internal app, which calls each capability as needed.
Are sales teams really building their own tools?
Many are starting to. Retool found 35% of teams have replaced at least one SaaS tool with a custom build, and Fullcast found 65% of senior revenue operators have postponed a purchase to build instead. Most still buy their system of record and specialist data.
What is MCP and why does it matter for sales?
MCP, the Model Context Protocol, is a standard way for AI assistants like Claude and ChatGPT to call outside tools. Salesforce, HubSpot, Apollo and 6sense all ship MCP servers, so an assistant can read and act on sales data without a person switching tabs.
What is the difference between an API and a CLI?
An API lets one piece of software call another and get structured data back. A CLI, or command line interface, lets a person or an AI coding agent run that capability by typing commands in a terminal. Most CLIs are thin wrappers around the vendor's API underneath.
Should we build or buy sales tools in 2027?
Usually both. Buy the capabilities that depend on proprietary data or hard judgment, such as CRM, contact data and relationship data, and build the surface your team works in. Assign an owner for maintenance and governance before anyone ships a custom tool.




