MCP and Skills in MACH (Short Read)
MCP and Skills in MACH: Access Is Not the Hard Part
The hardest part of MACH delivery is not calling vendor APIs. It is connecting vendors through bespoke integration, configuring each system to fit the business, and maintaining the glue between them. A five-system composable stack — CMS, PIM, OMS, DAM, OIDC — compounds this across every layer.
The cost of composable delivery is defined by three core layers:
- Architectural planning — which integration patterns, which consistency model, how to handle distributed transactions across services
- Vendor configuration — product data modeling in the commerce platform, content types in the CMS, auth flows in the identity provider. This is substantive delivery work, not toggle-flipping. Each vendor has its own modeling paradigm, and getting the structure wrong is expensive to fix post-launch.
- Bespoke integration code — the middleware, data mapping, and event pipelines that connect it all. Each arrow in a service diagram is an API call or event that must work reliably at scale.
These in turn drive two downstream costs that scale with the number of vendors:
- Training — business teams must learn to operate across multiple admin interfaces, each with its own concepts and workflows
- Support — dedicated runbooks per vendor, per integration point, maintained against ongoing API and feature changes
34% of brands attempting composable commerce report lacking the required vendor-specific expertise. [3] The question is whether AI agents — through MCP and Skills — can reduce the cost of any of these layers.
What MCP solves
MCP (Model Context Protocol) standardises how AI agents connect to vendor systems. CommerceTools exposes 95+ tools through its MCP server. Stripe provides 30+. Agents can query live vendor data through natural language instead of bespoke API integration per tool.
This is real progress — for the access layer.
What MCP does not solve
Cross-vendor orchestration. MCP standardises access to one vendor’s API. A multi-vendor stack needs separate MCP servers for each, with no mechanism for coordinating across them — no service discovery, no cross-server sequencing, no partial failure recovery. Planning decisions. MCP does not help you decide how to model your product catalog, whether to use event-driven vs. request-driven communication, or how to handle eventual consistency between a commerce platform and a search index. Configuration-as-delivery. MCP can execute API calls against a CMS, but it cannot decide what content types to create, how product attributes should be structured, or which auth flow fits the business requirements. Maintenance. When a vendor API change breaks downstream consumers, MCP does not detect or fix the cascade. The median enterprise runs 10 upgrade projects annually, with upgrade delivery consuming over a third of IT team time.
Where Skills fit
Agent Skills are a packaging format for vendor-specific procedural knowledge — instructions, templates, scripts, and examples that agents can discover and load on demand. Published by Anthropic in December 2025, the format is adopted across 30+ agents.
Skills are structured documentation, not magic. The value is making vendor-specific guidance discoverable by agents at task time, rather than depending on the developer to find the right docs page or the model to infer it from training data.
Where they help most:
- Vendor configuration guidance. How to model product types for a B2B catalog. How to structure content types in a CMS for a multi-locale storefront. How to configure organisations for multi-tenant access. Decisions that currently live in docs and consultants’ heads.
- Integration pattern selection. How to sync PIM products to a search vendor. How to map content models to product catalog structures. Which pattern for which combination.
- Cross-vendor sequencing. Set up the content model, create matching product types, configure the webhook bridge, test the sync. MCP gives access to each system individually; a skill gives the agent the sequence.
- Upgrade procedures. Which fields changed, which code patterns need updating, what tests to run after migration.
Where they do not help: architectural planning that requires business context and senior human judgment, organisational governance across a multi-vendor stack, and the full cross-vendor orchestration problem — sequencing across three or four vendors with partial failure handling remains bespoke engineering.
The vendor landscape
Most MACH vendors have shipped MCP servers: CommerceTools, Contentful, Contentstack, Auth0, Clerk, Kinde, Stripe, Talon.One, Cloudinary, Algolia, plus infrastructure providers Google Cloud and AWS and many more…
Fewer have packaged opinionated implementation guidance. Auth0 has shipped both MCP and Agent Skills. Clerk’s MCP server already delivers SDK snippets and code patterns alongside API access. The rest are mostly shipping access without structured workflow guidance.
The vendors most exposed are those in high-configuration domains — where product data modeling, content modeling, or auth flow configuration is the delivery bottleneck. These are the vendors where packaging implementation guidance would most reduce cost and error rates. Whether it arrives as Skills, rich MCP resources, or another format matters less than whether it arrives at all.
What remains unsolved
The biggest gap is cross-vendor orchestration. Neither MCP nor Skills solve the coordination problem across a multi-vendor stack. MCP talks to vendors individually. Skills encode workflows per vendor or per integration pair. The full orchestration layer remains custom engineering.
The bottom line
The hard part of MACH delivery is not API access — it is the planning, configuration, and integration work that consumers time, resources and money. MCP solves a real but narrow problem: standardised access to individual vendor APIs. Skills can address some of what MCP cannot — particularly vendor configuration guidance and integration sequencing in domains where expertise is scarce and mistakes are expensive. Neither solves the full problem. The vendors that package implementation guidance — in whatever form — will reduce the most expensive part of composable delivery.