Skip to content

03Protocol engineering

Model Context Protocol servers

Custom MCP servers enabling an AI assistant to perform CRM operations directly, across multiple company accounts, under per-user authentication.

Role
Backend & protocol engineer
Period
2025 — present
MCP servers in production
2
tools on the conversation server
18

Multi-tenant federation

Groups operating several brands maintain separate CRM accounts, each with distinct credentials and record schemas. Querying across them conventionally requires accessing each account individually.

The centralizer federates multiple Zoho tenants behind a single Model Context Protocol surface, allowing an AI client to query and act across every account within one session, with per-user token authentication constraining each operator to their own permissions.

AI clientClaudeMCP centralizerper-user token authTenant AZoho CRMTenant BZoho CRMTenant CZoho CRM
One Model Context Protocol surface federating several CRM tenants, with per-user tokens keeping each person inside their own permissions.

Conversation-layer tooling

A second server exposes eighteen tools over the conversation layer, covering thread search, history retrieval, agent assignment, labelling and reply dispatch.

Together these servers allow an assistant to execute CRM and messaging operations directly rather than describing them. SafqaiMCP is published openly on GitHub.

Two production MCP servers, eighteen tools, per-user token authentication across federated tenants.

Built with

  • Model Context Protocol
  • Claude
  • TypeScript
  • Zoho CRM
  • OAuth2