Skip to content
Skip to main content
PLATFORM/THE PROTOCOL/IMPULSE INTELLIGENCE PLATFORM
FIG. 00 · ONE INTERFACE
PLATFORM · THE PROTOCOL

Impulse Intelligence Platform

Most companies build integrations one at a time. They're about to be buried under their own technical debt.

90 minutes. No slides. Just the map.

One auth. One rate limiter. One audit log. One place to govern.

The standard pattern in 2024 was "build a Salesforce integration, then a HubSpot integration, then a Slack integration, then a custom one for the partner that just signed." Each integration had its own auth, its own rate limits, its own observability gap, its own breaking-change risk, its own dedicated maintenance.

That pattern was already getting expensive. In 2026, with AI agents now wanting to call into your kernel from Claude, ChatGPT, Perplexity, internal Breeze workflows, customer portals, partner systems, and a half-dozen other surfaces, the cost has compounded. Companies wiring integrations one at a time are now spending operational headcount on integration debt that should have gone into product or growth.

companyMCP (the Model Context Protocol that connects every interface to one intelligence layer) is what we install instead. Build the interface once. Every agent or tool that speaks MCP can call it.

This is not a productivity argument. It's a survivability argument.

A Model Context Protocol surface, tuned to your specific data model and your governance requirements.

In plain language: we stand up a single, governed interface that lets any caller (your team's AI tools, your customer-facing agents, your partner systems, a vendor's MCP client, your own internal Breeze workflows) ask your kernel a question and get a real, structured answer.

  • No prompt-stuffing.
  • No hallucinated numbers.
  • No prayer.

For most clients, the Kernel underneath is HubSpot — and other CRMs where they're already in place, though HubSpot is our preferred substrate. Depth built since 2007 pays back in protocol design.

Five categories. The exact shape is yours.

A typical companyMCP install includes five categories of endpoint. We design the shape with you in the diagnostic and proof-of-concept moves — not for you in a vacuum.

CATEGORY · 01
READ

Read endpoints

  • Contact & company queriesLookup by ID, by property, by association. Filtered by the access scope of the calling agent.
  • Deal pipeline queriesStage, amount, owner, age-in-stage, custom-object associations. Real-time.
  • Custom object queriesWhatever your kernel models. matters, engagements, projects, jobs, cohort records.
  • Workflow state queriesWhere this contact sits in onboarding. Whether a deal is stuck.
  • Reporting · natural language"How many enterprise deals are open in EMEA, past 30 days in negotiation, no activity in the last week."
CATEGORY · 02
WRITE

Write endpoints

The actions you authorize agents to take.

  • Notes · tasks · propertiesLifecycle stage transitions, deal stage moves, custom object creation.
  • Always governedBy your write policy. Versioned. Reversible.
CATEGORY · 03
SEARCH

Search endpoints

Natural-language search across the kernel.

  • Conversational queries"Find me the customer that mentioned procurement issues in their last call."
  • Indexed · structuredReturned with structured results, not pasted into a chat.
CATEGORY · 04
RESOURCES

Resource endpoints

The supporting data your agents need to be useful.

  • Brand voice rulesFrom your prompt library.
  • Approved messagingBy audience.
  • Recent customer conversationsKnowledge base articles. Compliance guidance.
CATEGORY · 05
TOOLS

Tool endpoints

The workflows you authorize agents to invoke.

  • Real workflows"Send the renewal pitch." "Open a partner deal registration." "Generate the proposal." "Escalate to CSM."
  • Arguments · validation · rollbackEach tool wraps a real workflow with all three.

A sales agent. Two real answers. One audit log entry.

A sales agent (running in Claude, or in your CRM, or from a Slack DM, depending on where the rep is working) needs to know which deals are at risk this quarter.

REQUESTThe agent calls →
Tool: company.search
Arguments: {
  "object": "deal",
  "filters": {
    "stage":             "negotiation",
    "days_in_stage":     ">30",
    "last_activity_days": ">7",
    "amount":            ">50000"
  }
}
  • 01Authenticate the caller against the rep's scope
  • 02Check authorization for this specific operation
  • 03Run the query against the kernel
  • 04Apply governance output filters
  • 05Log the call · return the result
RESPONSEThe protocol returns →
[
  {
    "deal_id": "1234",
    "name":    "Acme Renewal",
    "amount":  240000,
    ...
  },
  {
    "deal_id": "5678",
    "name":    "Beta Expansion",
    "amount":  180000,
    ...
  }
]
The agenthas structured data to reason over.
The repgets a real answer.
The kernelstate is up to date.
The audit loghas the trail.
× No prompt-stuffing.× No screenshot pasting.× No hallucinated numbers.

The protocol layer is also where governance lives.

This is the layer that lets you put an agent in front of a customer without losing sleep.

AUTH
Authentication
OAuth or signed tokens. Scoped to the calling agent and the specific operation. Token rotation built in. Revocation immediate.
SCOPE
Access scope
The calling agent sees only what it's authorized to see. Customer-facing agents see only that customer's data. Partner-facing agents see only their deal registrations. Internal sales co-pilots see the rep's territory.
AUDIT
Audit log
Every call. Every actor. Every operation. Every response. Every timestamp. Retained per policy. Searchable. Exportable for compliance audits.
LIMITS
Rate limits
Per agent. Per scope. Per operation. Tunable. Alerts on anomalies.
OUTPUT
Output controls
Schema enforcement on responses. PII redaction where required. Brand-voice consistency checks where the response will be customer-facing.
ROLLBACK
Rollback
Write operations are versioned. Mistakes are reversible.
Agent Governance and Audit (services) →

Seven callers. One surface.

01INTERNAL
HubSpot Breeze
For agents running natively inside the kernel — and Breeze is just the start of what's possible once your kernel is exposed.
02TEAM
Claude
When team members use their personal AI to do work that touches the kernel.
03TEAM
ChatGPT
For the same pattern in teams standardized on OpenAI.
04OWN
Custom agents
We've built — running inside the Surface layer or on partner integrations.
05CUSTOMER
Customer-facing portals
Conversational UI for self-serve workflows.
06PARTNER
Partner systems
Exposing a controlled slice of the kernel for deal registration or lifecycle visibility.
07OWN
Internal tooling
Your engineering team builds against the same surface.
Custom MCP Servers service →

Four things, just to save the procurement team time.

× IT IS NOT
A public REST API with a Swagger doc.
IT IS
An MCP surface built for agent consumption — not for general-purpose developer integration.
× IT IS NOT
A single SaaS product you buy and turn on.
IT IS
An installed protocol surface, designed to your kernel and governed to your risk tolerance.
× IT IS NOT
A replacement for HubSpot's own API.
IT IS
A complement that sits alongside it. The HubSpot API stays available for traditional integrations; the MCP surface is for agent-mediated access.
× IT IS NOT
Insecure-by-default.
IT IS
Governed from day one. Auth, scope, and audit infrastructure are core to the install — never an afterthought.

Honest read. Two columns.

The diagnostic determines which case you're in. We won't sell you protocol installation you don't need.

YOU NEED THIS WHEN
  • You're deploying multiple agents (your own, your team's, partner-built) and don't want to maintain N separate integrations.
  • You have customers or partners who should have agent-mediated access to your kernel.
  • You're at L3 or higher on the Maturity Model and ready to move toward L4 Exposed.
  • Your governance and compliance posture requires a single audited interface for agent-mediated access.
×YOU DON'T NEED THIS WHEN
  • You're at L2 or below. Build the Kernel first. The protocol on top of a broken kernel produces broken agent responses faster.
  • Your AI use is exclusively single-agent inside HubSpot's Breeze ecosystem. Native Breeze tooling is sufficient for that case.
  • Your roadmap doesn't include customer-facing or partner-facing agent surfaces.
DIAGNOSTIC · 90 MINUTES · NO SLIDES

Map your company's intelligence in 90 minutes.

We will assess your kernel's readiness for protocol exposure during the 90 minutes. You walk away with a written read on whether MCP installation is the right next move — or whether kernel work has to come first.

Yours to keep — signed or not.