The SaaSpocalypse, MCP Servers, and What Service Businesses Need to Understand Right Now

Turning service business expertise into scalable software products.

$285 billion wiped from SaaS stocks in a day. MCP servers growing 80x in 12 months. Here's what the data says about where value is shifting — and why service businesses should be paying attention.

The SaaSpocalypse, MCP Servers, and What Service Businesses Need to Understand Right Now

On January 30, 2026, Anthropic launched 11 open-source plugins for Claude Cowork. Within hours, $285 billion in SaaS market cap evaporated. Salesforce, ServiceNow, HubSpot, Monday.com — all dropped. The financial press started calling it the "SaaSpocalypse."

I've spent weeks researching what actually happened, what's driving it, and what it means for people who run service businesses. Not because I had a strong thesis going in, but because the scale of the market reaction suggested something structural was shifting — and I wanted to understand it properly.

Here's what I found.

---

The selloff wasn't random. It targeted specific categories.

Jefferies coined the term "SaaSpocalypse." The Goldman Sachs Software Index fell 30% from its October 2025 highs. Software price-to-sales ratios compressed from 9× to 6×. But the damage wasn't spread evenly.

Customer support tools took the hardest hit. Claude Cowork shipped a dedicated customer support plugin. AI startup Sierra reached $100M ARR in two years competing directly with incumbents. Freshworks dropped 36.8% year-to-date — its IPO investors have lost 85% of their investment. Intercom's Fin AI agent already resolves 65% of conversations end-to-end without a human.

Project management was what KeyBanc analyst Jackson Ader called "most exposed." Monday.com fell 20% in a single session despite beating earnings. Asana collapsed 59% over 12 months and now trades at 3× ARR — down from 89× at its 2020 peak. The logic: seat-based tools not tied to a deep system of record are vulnerable because AI agents don't need dashboards to manage tasks.

SMB CRM faces the starkest maths. If AI agents reduce a 10-person sales team to 2 people, that company needs 2 CRM seats, not 10. HubSpot dropped 39% year-to-date. Salesforce fell 25-43% and sits below 4× revenue. As SaaStr's Jason Lemkin put it: "AI isn't eating the product. It's eating the budget."

Marketing automation is under direct pressure from Cowork's marketing plugin for content creation, campaign analysis, and SEO. Wells Fargo downgraded Intuit from "Overweight" to "Equal Weight."

Document management and collaboration round out the top five. Box fell 17% year-to-date. Figma dropped 40%.

The categories that are holding up tell you where the moat still exists: cybersecurity (more AI agents means more attack surface — the market actually grows), ERP systems (too deeply embedded to replace — AI agents use ERPs, they don't displace them), and accounting software (regulatory moats around tax filing and compliance).

The pattern is clear. Generic software that competes on UI and feature count is getting repriced. Software built around deep domain logic, proprietary data, and regulatory expertise is holding its value.

That distinction matters a lot for service businesses.

---

What MCP actually is (in plain English)

Model Context Protocol is the open standard that lets AI tools connect to external data and services. Anthropic launched it in November 2024. Think of it as the USB-C of AI — a universal port that lets any AI assistant plug into any data source.

The adoption has been extraordinary. OpenAI integrated MCP in March 2025. Google's Demis Hassabis called it "rapidly becoming the open standard for the AI agentic era." Microsoft joined the steering committee at Build 2025. In December 2025, Anthropic donated MCP to the Linux Foundation's Agentic AI Foundation, co-founded with OpenAI and Block, with AWS, Microsoft, Google, Bloomberg, and Cloudflare as supporting members.

The numbers: 97 million monthly SDK downloads. Over 17,000 MCP servers listed on mcp.so. An 80× increase in SDK downloads in the first five months alone.

In practical terms: when someone builds an MCP server, they're making a data source or service accessible to AI assistants. A recruitment database becomes something Claude can search. A compliance framework becomes something AI can apply. An assessment methodology becomes something clients can query through natural language.

The reason this matters beyond developer tooling is that MCP turns proprietary methodology into infrastructure. Not a dashboard someone logs into. Infrastructure that AI tools can call directly.

---

February 2026: the enterprise wave arrives

Since this article was first published, the MCP ecosystem accelerated faster than even the bullish projections suggested. February 2026 brought a cluster of enterprise launches that moved MCP from "developer infrastructure" to "business operations layer":

Atlassian Rovo MCP went generally available on 4 February, giving Claude full read and write access to Jira and Confluence for 16+ AI clients. This isn't a read-only integration — teams can create Jira tickets, update project statuses, and search Confluence documentation entirely through AI conversation.

Amazon Ads MCP entered open beta on 3 February, letting advertisers manage campaigns through natural-language conversation. Create campaigns, adjust budgets, pull performance data — all through Claude or any MCP-compatible client.

WordPress launched an MCP Adapter on 4 February built on its Abilities API. Millions of WordPress sites can now be managed through AI conversation.

Google launched a Developer Knowledge MCP server on 13 February, making its documentation machine-readable for AI clients.

Supermetrics added MCP support for its marketing data platform, connecting Claude to analytics from 150+ marketing platforms.

Perhaps most telling: paid MCP servers have emerged at price points from $300 to $6,000 per month for enterprise-grade integrations. A pricing layer forming this quickly signals real commercial demand, not speculative interest.

Each of these launches reinforces the same pattern: MCP is becoming the standard interface between AI and business tools. The businesses that have their methodology accessible through this interface benefit. Those that don't become invisible to AI workflows.

For a detailed guide to which MCP servers matter most for service business owners, see The Best MCP Servers for Business Owners.

---

Almost nobody outside dev circles knows about this yet

Here's the finding that surprised me most. Developer awareness of MCP is at saturation — it's on the Thoughtworks Technology Radar, Microsoft has published tutorials, every major AI tool supports it.

Drop outside that technical bubble and awareness falls off a cliff.

I found a handful of accessible resources aimed at non-technical audiences: Greg Isenberg covered MCP on the Startup Ideas Podcast, TrueFuture Media published a marketing-focused explainer, and HubSpot co-founder Dharmesh Shah wrote a viral post calling MCP "the next billion-dollar AI idea."

That's about it. No "MCP for Business" newsletter exists. No industry-specific guides. No webinar series for non-technical founders. No mainstream business press coverage (no Forbes, Inc., or Entrepreneur articles on MCP for business use cases).

No dedicated "MCP for Recruitment Firms" or "MCP for Training Providers" or "MCP for Compliance Consultancies" content exists anywhere — despite clear applicability to all three.

The awareness gap between developers and business owners is enormous. And if the SaaSpocalypse data tells us anything, it's that the gap won't last long. Every headline about AI replacing SaaS pushes more business owners toward understanding what's actually happening.

---

The MCP services market: crowded at the bottom, empty at the top

I mapped every company, agency, freelancer, and platform globally that offers MCP-related services. The picture is revealing.

At the commodity end: 24+ freelancers on Fiverr offer basic MCP server builds for $90-$100. Indian IT services firms have published SEO-optimised service pages. Mid-range agencies charge $25,000-$50,000 for SMB implementations.

Dedicated MCP agencies are rare. mcp-agency.com in Germany offers packages from €5,000-€30,000. Agency (agen.cy) targets Fortune 500 implementations. Advisor Labs focuses on legacy system integration.

SaaS platforms that help you build MCP servers are emerging fast. MCP-Builder.ai offers no-code server generation for $30-$225/month. Composio provides 500+ pre-built integrations. Smithery.ai hosts nearly 3,000 MCP servers.

Y Combinator dedicated roughly half of its Spring 2025 batch — 70+ of 144 companies — to agentic AI. Gartner forecasts 33% of enterprise software will include agentic AI by 2028, up from under 1% in 2024.

But here's what's missing: almost all of this activity is about connecting existing tools to AI, or building developer infrastructure. The specific application of taking a service business's proprietary methodology — the thing a consultant or specialist actually does that makes them valuable — and making it accessible through MCP is essentially unaddressed.

The companies building MCP servers are building them to connect Slack, GitHub, databases, and cloud services. They're not building them to encode how a specific recruitment firm matches candidates, or how a specific compliance consultancy scores risk, or how a specific training provider assesses readiness.

That gap represents something important.

---

The layoff pattern tells the same story

While SaaS stocks were falling, something equally significant was happening in professional services. Baker McKenzie — one of the world's largest law firms with $3.4 billion in revenue — announced the elimination of 600 to 1,000 support staff roles, citing AI as the primary driver. The cuts target research, marketing, know-how, and secretarial functions. Not lawyers — support staff.

This follows Clifford Chance cutting 50 support roles and Freshfields replacing paralegals with AI systems. Across industries, over 22,000 AI-driven layoffs have been announced in 2026 so far, following 54,000+ in 2025.

The pattern is consistent: AI isn't replacing the expertise. It's replacing the manual processes around the expertise. The lawyers at Baker McKenzie still have their jobs. The people who did research, formatting, and document preparation do not.

This is precisely the dynamic that creates opportunity for service businesses. The support functions being eliminated — research, analysis, document preparation, matching, scoring — are the same functions that can be encoded as MCP tools and sold as product. The 600+ people leaving Baker McKenzie carry methodology that could be productised. The recruitment support staff at Salesforce (which cut roughly 1,000 people in February 2026), the marketing teams at McKinsey (which cut 200 tech staff) — all of them carry domain-specific process knowledge.

When Forrester Research found that 55% of companies regret their AI-driven layoffs and 50% plan to rehire offshore at lower salaries, it revealed something important: the methodology can't simply be deleted. It has to go somewhere. The question is whether it goes to cheaper humans or gets structured as product.

---

Why the value shift matters for service businesses specifically

The SaaSpocalypse logic boils down to this: if AI becomes the primary interface for work, generic UI-layer software loses value. What retains value is the intelligence underneath — proprietary data, domain-specific logic, and expert methodology.

Service businesses have been sitting on that intelligence for years. Decades, in some cases.

A recruitment firm's candidate matching methodology. A compliance consultancy's risk scoring framework. A training provider's competency assessment model. An HR consultancy's onboarding process. A marketing agency's campaign evaluation system.

These are all examples of proprietary intelligence that currently lives in spreadsheets, in people's heads, or in manual processes. They're the reason clients pay for the service. And they're exactly the type of intelligence that becomes more valuable — not less — as generic software gets commoditised.

The question isn't whether this intelligence has value. It obviously does. The question is whether it's structured in a way that makes it accessible beyond the individual humans who currently carry it.

That's what MCP enables. Not replacing the people. Making their methodology available at 2am on a Tuesday when no human is working. Making it queryable through AI tools that millions of people already use.

---

Claude's trajectory puts platform risk in context

When people ask whether this platform is real enough to build on, the numbers answer the question.

Anthropic hit $14 billion in annualised run-rate revenue as of February 2026. Up from roughly $1 billion in 2024. The Series G raised $30 billion at a $380 billion valuation — the second-largest private tech financing in history.

Eight of the Fortune 10 are Claude customers. Over 500 companies spend more than $1 million annually, up from 12 two years ago. Claude Code alone generates over $2.5 billion in annual revenue and accounts for 4% of all public GitHub commits worldwide.

30 million monthly active users. 18-minute average sessions. 41% return three times a week.

80% of consumer Claude usage comes from outside the United States. Japan is Anthropic's top international priority — APAC revenue has grown over 10× in the past year. South Korea has higher per-capita Claude usage than the US. In Europe, Anthropic opened offices in Munich and Paris in late 2025.

Claude Cowork — the non-developer interface — is available on Mac and Windows for Pro, Max, Team, and Enterprise subscribers. It's positioned as "Claude Code for the rest of your work."

This isn't speculative infrastructure. Tens of millions of people are spending 18 minutes a day in this environment doing real work, three times a week.

---

International signals: this is moving everywhere

China's Big Five tech companies (Alibaba, Baidu, Tencent, ByteDance, Zhipu AI) are all deploying MCP-based services. Alibaba Cloud launched an MCP marketplace with 1,000+ services. IDC projects China's AI agent market will surge 75× by 2028.

South Korea's Kakao launched an MCP framework with a developer marketplace. SK Telecom's AI service has 10 million monthly active users committed to MCP.

In Europe, the SAP-Mistral AI partnership is building autonomous agents using MCP-compatible frameworks. A sovereign AI partnership between France, Germany, Mistral, and SAP will create AI agents for public administration by mid-2026. The MCP Developers Summit Europe in London sold out.

Japan's severe workforce shortage creates particularly urgent pull for AI automation — Rakuten reported a 79% reduction in feature development time with Claude Code.

The standard is genuinely global. MCP adoption isn't limited to Silicon Valley or English-speaking markets. Any methodology productised through MCP has a potential global audience from day one.

---

Claude Skills add another layer

In October 2025, Anthropic launched Claude Skills — a way to teach Claude how to use tools the way a specific team actually works. Developer Simon Willison called Skills "maybe a bigger deal than MCP," predicting a "Cambrian explosion in Skills."

The distinction matters: MCP gives Claude access to tools and data. Skills teach Claude how to apply methodology to that data. Together, they create something new — AI that doesn't just retrieve information from your system, but applies your framework when answering questions.

A recruitment firm's MCP server makes their contractor database queryable. Their Claude Skill applies their matching methodology when ranking results. The combination means a hiring manager can ask "Who's the best fit for this role?" and get an answer that reflects the firm's actual expertise — not generic AI reasoning.

That combination of data access and methodology application is difficult to replicate. It's not a feature someone copies. It's an encoding of accumulated expertise.

---

The window and what closes it

Three forces will eventually close the current window of opportunity:

No-code platforms are improving monthly. MCP-Builder.ai already lets people generate basic MCP servers for $30/month. Today they're primitive — they connect existing APIs, not proprietary methodology. In 18 months, they'll be significantly more capable.

Venture capital is flooding the space. Y Combinator is funding 70+ agentic AI companies per batch. At some point, venture-backed startups will start targeting specific verticals — recruitment, compliance, training — with polished products that offer methodology-as-a-service for specific domains.

The awareness gap is shrinking. Every SaaSpocalypse headline, every Dharmesh Shah post, every podcast episode about MCP pushes more non-technical business owners toward understanding what this enables. The February 2026 enterprise launches from Atlassian, Amazon, and WordPress move MCP further into mainstream business consciousness.

The businesses that structure their methodology and make it accessible now will have live systems, real usage data, and client feedback before their competitors understand what MCP stands for.

The businesses that wait will be responding to competitors who moved first.

---

What this means practically

I'm not going to pretend this is a simple "just build an MCP server" situation. Structuring proprietary methodology into something AI can use is genuinely hard intellectual work. It requires understanding data modelling, product design, and domain expertise simultaneously.

But the core question for any service business owner reading this is worth sitting with:

Is the methodology that makes your business valuable currently trapped in people's heads and spreadsheets? And what happens when someone in your space structures theirs first?

The SaaSpocalypse proved that generic software is being repriced. The MCP ecosystem proves the infrastructure is ready. The awareness data proves the market hasn't caught up yet. The layoff wave proves that methodology is already being displaced from organisations — the question is whether it gets structured as product or simply lost.

That combination — proven value shift, ready infrastructure, low awareness — is what a genuine first-mover window looks like.

Whether you build something yourself, hire someone to do it, or just start educating yourself on what MCP enables — the worst move is to ignore this entirely and hope the world stays the same.

It won't. The data is pretty clear on that.

---

FAQ

What is MCP?

MCP stands for Model Context Protocol. It's an open standard — originally created by Anthropic, now maintained by the Linux Foundation — that lets AI assistants connect to external data sources and tools. Think of it as a universal adapter: build an MCP server for your data, and any compatible AI tool (Claude, ChatGPT, and others) can access it.

What happened in the SaaSpocalypse?

On January 30, 2026, Anthropic launched 11 open-source plugins for Claude Cowork that could replace functions of major SaaS tools. Software stocks dropped sharply — $285 billion in market cap was wiped in a single day. The most affected categories were customer support, project management, CRM, and marketing automation.

Why does this matter for service businesses?

The SaaSpocalypse showed that generic UI-layer software is losing value. What retains value is proprietary intelligence — the methodology, frameworks, and domain expertise that service businesses have built over years. MCP provides the infrastructure to make that intelligence accessible through AI, creating a new product layer that didn't previously exist.

Do I need to be technical to understand MCP?

No. The concept is straightforward: MCP lets AI tools access your data and apply your methodology. The implementation requires technical skills, but the strategic decision — whether to structure your methodology and make it accessible — is a business decision, not a technical one.

Is MCP only for Claude?

No. While Anthropic created MCP, the protocol is now supported by OpenAI, Google, Microsoft, and others. It was donated to the Linux Foundation in December 2025. Building on MCP means building for every major AI platform, not just one.

How established is the MCP ecosystem?

Very. 97 million monthly SDK downloads. Over 17,000 servers listed on mcp.so. Supported by every major AI company. Used by companies from startups to Fortune 10 enterprises. February 2026 brought GA launches from Atlassian and Amazon, new servers from WordPress and Google, and the emergence of paid enterprise MCP servers. The ecosystem went from near-zero to industry standard in about 14 months.

---

Tom Crossman builds products for service businesses at Hello Crossman. 18 years in product development. Head of Product Engineering at Habito (£3B in mortgages processed). 100+ products shipped.

---