This is the closing of the MCP series. Twenty-four articles on what MCP is, how to build with it, what disciplines apply. The closing is forward-looking.
What's coming
Multi-modal arguments. MCP currently handles text. Images, audio, structured data are coming.
Better discovery. Tool registries that make finding existing MCP servers easier.
Authorisation standards. Per-user, per-resource, fine-grained.
Performance optimisations. Streaming responses, partial results, caching.
Cross-server composition standards. Patterns for server-to-server with guarantees.
These are realistic 1-2 year horizons.
What to bet on
For teams investing in MCP:
- The protocol will stabilise.
- The ecosystem will grow.
- Migration costs from non-MCP-compatible tools will rise (sticky cost of legacy).
- Cross-provider portability will improve.
The investment in MCP-compatible tooling is a durable bet.
What to wait on
- Multi-modal MCP (still maturing).
- Some authorisation patterns (in flux).
- Some hosted-service patterns (vendor experimentation).
For these, build pragmatically with current MCP and migrate when standards land.
A real prediction
Within two years, most enterprise SaaS tools will expose MCP servers. The teams that built MCP-first will integrate easily. The teams that didn't will face migration overhead.
Limits
Predictions in this space are unreliable. The protocol may evolve in unexpected ways. New entrants may displace MCP. The team's bet should be on the principles (standardisation, composability, discoverability) more than on the specific protocol.
Close
This concludes the MCP series. The protocol is the standard. The disciplines covered apply. The future is multi-modal, more discoverable, more composed. Build now with current MCP; the investment compounds.
This also concludes the 200-article curriculum across eight series. From AI agents in industry to MCP and AI protocols, the curriculum maps the territory of building AI-enabled software in 2026 and beyond.
The next article — when it lands — will be a new series. The disciplines covered here are foundational; the depth ahead is deep.
Related reading
- MCP in 10 minutes
- Why we need MCP at all — foundational framing.
- Browse all series — the full curriculum map.
We build AI-enabled software and help businesses put AI to work. If you're betting on MCP, we'd love to hear about it. Get in touch.