Service pillar

MCP server development for WordPress and WooCommerce

Senior B2B, EU jurisdiction, scoped per project.

Pricing is individual. I reply within one working day.

What I ship

  • Custom Anthropic MCP server in TypeScript, deployed on Cloudflare Workers or a Node runtime, exposing typed tools for the catalogue and order-intent flows the client needs.
  • Schema.org enrichment in parallel, so the same catalogue stays legible to crawlers and AI assistants that do not yet speak MCP. Schema.org and the MCP tool surface stay in sync as the catalogue evolves.
  • Observability for agent traffic separated from human traffic at the analytics layer: per-tool latency, error rate, authorisation failures, and conversion attribution.
  • Claude API test harness that exercises every MCP tool end to end against a staging WordPress, with recorded transcripts as the regression baseline. Shipping a new tool requires a green run on the harness.

When MCP is the right choice

An MCP server earns its place when an agent needs to take action rather than only read. The action might be creating a B2B quote with custom pricing, applying a partner discount code, changing a subscription cadence, or committing to an order on a logged-in shopper's behalf. The interface for those actions has to be typed, authenticated and idempotent; that is the contract MCP gives you.

The second case is authentication. When the agent is acting inside a shopper session, on behalf of a partner, or under a machine-to-machine credential, the catalogue surface needs to know who is calling and what they are allowed to do. A typed MCP tool with an authorisation policy is the right shape for that; a public REST endpoint is not.

The third case is stateful flows. Carts, quotes, subscriptions and order intents have lifecycle. Determinism and idempotency keys matter so that a retried call does not double-charge or duplicate a quote. MCP tools express that intent explicitly; scraping the storefront does not.

Who this is for

  • WooCommerce stores with custom workflows: subscriptions, B2B quotes, partner discounts, multi-region pricing, custom checkout steps.
  • B2B catalogues where every transaction needs an authenticated quote rather than a self-service add-to-cart, and where the buyer is increasingly an agent acting on behalf of a procurement team.
  • Content sites with affiliate commerce, where the page is the catalogue and an MCP tool surface lets agents reason over Offers, availability and partner terms in one structured call.
  • Headless WordPress setups already running Astro or Next.js on Cloudflare Workers, where the MCP server slots in alongside the existing front end without changing the editorial back end.

Engagement model

Senior B2B contracts on EU jurisdiction. Discovery, scoping, build and integration, tuning. Scope and pricing are individual per project; I do not publish standard rate cards because the discovery step usually changes the cost of the build.

Agents read the Schema.org layer for browsing and the MCP layer for transactional intent. WordPress remains the canonical source. AI agent (Claude, ChatGPT, browser) reads Schema.org Product + AggregateOffer for browsing and exchanges transactional intent via MCP server (typed intent). WordPress / WooCommerce origin holds the Canonical catalogue + orders. Webhook on price / stock change. AI agent (Claude, ChatGPT, browser) Schema.org Product + AggregateOffer machine-readable read MCP server (typed intent) transactional intent WordPress / WooCommerce origin Canonical catalogue + orders Webhook on price / stock change
Agents read the Schema.org layer for browsing and the MCP layer for transactional intent. WordPress remains the canonical source.
MCP provides a universal, open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol.
Anthropic , MCP launch announcement , Model Context Protocol , 2024-11-25 , source

Frequently asked questions

What is the Model Context Protocol?

Model Context Protocol (MCP) is an open standard published by Anthropic in November 2024. It defines how AI applications connect to external tools and data sources through a typed, authenticated interface. Instead of every model vendor inventing a bespoke integration shape, MCP gives both sides one contract to implement. For a WordPress or WooCommerce store it means an agent can call typed tools (lookup product, check stock, create quote, signal order intent) rather than scraping the public site.

When does a custom MCP server make sense vs Schema.org enrichment alone?

Schema.org Product, Offer and AggregateOffer markup is the right starting point for any catalogue. It is necessary but not sufficient. A custom MCP server is the right next step when the agent needs to take action rather than just read: create a B2B quote, change a subscription, apply a partner discount, or commit to an order. It is also justified when authentication context matters, when state is involved, or when WooCommerce custom flows do not map cleanly onto crawlable JSON-LD.

How does authentication work in an MCP server?

Two common patterns. OAuth on behalf of an end user, where the agent acts inside an authenticated shopper session and the MCP server enforces the same permissions WordPress would. Personal access tokens or signed machine-to-machine credentials, where a partner agent identifies itself and is granted a scoped capability set. I also support per-tool authorisation policies, idempotency keys for mutating tools, and full request audit logs.

What is the cost model for an MCP engagement?

Senior B2B contracts on EU jurisdiction, scoped per project. Pricing is individual; I do not publish standard rate cards because the discovery and scoping steps usually change the cost of the build. Runtime cost on Cloudflare Workers is per request rather than per server, which keeps the operational surface predictable as agent traffic grows.

When is MCP not the right choice?

When the catalogue is small, fully static, and the desired agent behaviour is read-only. In that case disciplined Schema.org enrichment plus a clean WordPress REST surface is faster, cheaper and sufficient. MCP earns its place when there is a typed action to take, an authentication boundary to enforce, or a stateful flow (cart, quote, subscription change) where determinism and idempotency matter.

Proof through AI governance

MCP work should start with internal AI governance: tool boundaries, permissions, regression prompts, logging and rollback. The anonymised AI content-ops case study shows that operating model before it becomes a public agent surface.

Cluster reading

Adjacent service pillars

MCP cluster supporting articles

Agent-readable web and AI integrations

Reference

Start an MCP engagement

Tell me the catalogue, the agent flows you want to support, and the timeline. I reply within one working day.

Contact me