Skip to content
Start a conversation
← All articles
AIAgents

AI already connects to everything. Your operations, not yet

MCP is now the standard for connecting AI to software. In Mexico, the disadvantage isn't the model. It's the context AI can connect to.

Armando Valenzuela Williams6 min readUpdated
Pintura abstracta de acrílico fluido en azul, naranja y negro

Half of the companies in Mexico already use artificial intelligence, but almost none have given it access to how their business actually runs. At Kirana Labs we work on that problem through Digital Enterprise, our service line for companies of 10 to 300 people that want to expand margin with better systems, well-designed processes, and real-time visibility. This article covers what MCP is, what the 2026 numbers say, where Mexican companies fall behind, and how we solve it with KAI Vault.

What MCP is

MCP (Model Context Protocol) is an open standard Anthropic released in November 2024 so AI assistants can query and operate other systems.

Before MCP, every combination of AI and software needed its own integration. Connecting ChatGPT to your ERP was one project, and connecting Claude to the same ERP was another. With MCP, a system publishes a server once and any compatible assistant can use it.

An MCP server tells the AI what information it can read, what actions it can take, and with which permissions. With that, the AI stops answering from memory and starts answering with your company's data: today's inventory, the status of an order, the quote that went out yesterday.

The 2026 numbers

MCP went from proposal to standard in under two years. In December 2025, when it moved to a neutral foundation under the Linux Foundation with OpenAI, Google, and Microsoft as co-sponsors, it had more than 10,000 active public servers and 97 million monthly SDK downloads.

Seven months later, in July 2026, downloads were approaching 500 million a month, and the TypeScript and Python SDKs had each passed 1 billion cumulative downloads. ChatGPT, Claude, Gemini, Microsoft Copilot, Cursor, and VS Code all support it.

In seven months, monthly downloads grew fivefold. MCP is no longer a technical bet. It's how AI connects to software.

Where companies in Mexico stand

AI adoption in Mexico grew fast. According to the Strand Partners study for AWS, it went from 38% to 48% of companies in 12 months, with small and mid-sized businesses at 47%. Banxico reports that among companies with more than 100 employees, the share doubled between September 2025 and June 2026, from 24% to 49%. In northern Mexico it sits at 46%.

But adoption is wide, not deep. 63% of companies remain in basic uses and only 13% have reached the most advanced stage. Among companies familiar with agentic AI, only 6% have fully implemented it. 46% have no reliable way to measure the return on what they've already invested in AI, and 36% cite technical or data challenges as a barrier.

Using ChatGPT to draft emails counts as adoption. It doesn't change how the company runs.

Where the disadvantage is

The disadvantage isn't the model. An 80-person company in Monterrey has access to the same AI as an 8,000-person company in the United States. The difference is the context that AI can reach, and three things hold it back.

Operations live where there's no connector. Salesforce, HubSpot, Google Drive, and Notion already publish MCP servers. A mid-sized company in Mexico runs somewhere else: WhatsApp groups, Excel files sent by email, a local ERP, and a system someone built ten years ago. An agent can't read what has no connector.

The data doesn't match. Sales, operations, and finance carry three different numbers for the same thing. Connecting AI to three versions of the truth produces three answers, and the CEO stops trusting all three.

Nobody decided what AI can see. Without a permissions layer, the choice is all or nothing, and most owners choose nothing. That isn't excess caution: in the same study, 52% of companies prioritize strong data protection standards and 46% want control over how and where their information is processed.

The result is that the agent that works in the demo arrives at the company and has nothing to read.

What KAI Vault is

KAI Vault is the Kirana Labs layer that brings a company's digital information into one place, with permissions, and exposes it to AI through a single MCP server. It solves the three problems above. It connects what has no connector: cloud systems, legacy systems, and communication channels like email and WhatsApp. It defines which data wins when two sources disagree. And it controls, by role, what the AI can read and do.

How it works

  1. Mapping. We identify what information exists, where it lives, who owns each piece of data, and which source is authoritative.
  2. Connection. We connect those sources to the Vault without replacing anything: the ERP, the CRM, spreadsheets, email, company channels, and legacy systems.
  3. Permissions. We define, by role, what the AI can query and which actions it can take.
  4. One MCP. The Vault publishes one MCP server, and Claude, ChatGPT, Copilot, or your own agents query it there. If you switch models tomorrow, you don't redo integrations.

With that, a question like "which Monterrey orders are running late and what did we promise the customer?" gets answered in one query instead of checking three systems and a chat.

Where it fits on the ladder

KAI Vault is the second of four stages for a company to run on AI:

  1. Data mapping and readiness.
  2. Vault: all digital information in one place.
  3. Analytics and KPIs on top of that information.
  4. Agents that execute.

Most companies try to start at stage four. That's why only 6% of those familiar with agentic AI have it running. Context first, agents second.

About Kirana Labs

Kirana Labs is a nearshore software development and automation agency based in Monterrey that helps companies expand margin, founders get to revenue, and technical teams let go of production with confidence.

If you want to know which stage your company is at, schedule a call.

Sources

Start a conversation.

Tell us the number you want to move: a launch date, a cost, a cycle time, a cash position. We'll come back with a baseline, a plan, and what we'd be accountable for.