In 2023, integrating an AI assistant with the team's tooling meant writing custom code per assistant per tool. Slack-Claude, Slack-ChatGPT, Linear-Claude, Linear-ChatGPT — the matrix grew quadratically.
This was the integration drawer: a place full of cables, each one specific to one device-pair, none reusable.
MCP is the USB-C of AI tooling. One protocol; many assistants; many tools.
The problem before MCP
A team building an AI integration faced:
- Per-assistant SDK choices.
- Per-tool authentication patterns.
- Per-team-member configuration.
- No standard transport.
- No schema standard for tools.
Each integration was custom work. Switching assistants meant re-doing integrations. Switching tools meant re-doing assistant configurations.
MCP's answer
MCP introduces:
- One protocol for tool description.
- One schema for tool arguments.
- Three standardised transports.
- One pattern for auth.
The team writes the integration once. Any MCP-compatible assistant can use it.
A real pain
A team integrated Slack with Claude in 2023. Custom code: 800 lines.
Same team integrated Slack with Claude in 2026 via MCP: 0 lines (used existing MCP server).
The MCP server itself is reusable across assistants. The team's investment in MCP-compatible tooling compounds.
How MCP helps
For teams:
- Faster integration (use existing servers).
- Lower lock-in (servers are portable).
- Better tooling ecosystem (vendors target MCP).
- Standards-based scaling.
For vendors:
- Build one MCP server, support many assistants.
- Less per-assistant maintenance.
- Wider distribution.
What we won't do
Avoid MCP because the team's AI assistant is "different." Most major assistants now support MCP.
Build custom integrations when an MCP server already exists.
Lock into a single AI provider when MCP makes the choice reversible.
Close
MCP solves the integration-drawer problem. Before, every integration was custom. Now, one protocol; many assistants; many tools. The team's investment scales. Skip MCP and you're rebuilding integrations every time the AI provider changes.
Related reading
- MCP in 10 minutes — primer.
- MCP servers are USB-C for AI — framing.
- Tool design like APIs — companion discipline.
We build AI-enabled software and help businesses put AI to work. If you're considering MCP, we'd love to hear about it. Get in touch.