MCP servers and agent workflows
Model Context Protocol servers that let LLMs work with OneDrive, Outlook and SharePoint through Microsoft Graph.
- Role
- Full Stack AI Engineer
- When
- Mar 2026 to present
- Status
- Public on GitHub
- Stack
- Node.js, Express, Microsoft Graph, Docker
Results
- Public on GitHub
- 3 serversOneDrive, Outlook, SharePoint
- SharePoint server
- 18 toolsSites, libraries, lists, sharing, search, file reading
The problem
LLMs are only as useful as the systems they can reach. At Ajaia I build MCP servers and AI integrations that connect models to external tools, APIs and business systems.
Approach
Three public servers cover Microsoft 365 through Microsoft Graph. Each is a Node.js and Express service with its own Dockerfile.
The SharePoint server exposes 18 tools across sites, document libraries, lists, permissions and sharing, search, and reading file contents from DOCX, PDF, XLSX, PPTX, TXT, CSV and JSON.
The Outlook server covers mail, attachments, calendar and contacts. The OneDrive server covers drives, files, search and users.
At Ajaia I also build reusable AI agent and MCP workflows.
Simplified, illustrative diagram
MCP servers and agent workflows
An LLM app or agent decides it needs a tool.
Public on GitHub
3 Microsoft 365 MCP servers
AI client
An LLM app or agent decides it needs a tool.
Key decisions
Readable results first, structured data alongside
The SharePoint tools return a user-friendly result first, with structured data next to it for downstream steps. The model gets something it can quote, and the workflow gets something it can parse.
One server per Microsoft 365 surface
OneDrive, Outlook and SharePoint are separate servers. Each declares its own Microsoft Graph scopes (file scopes for OneDrive, mail scopes for Outlook) and ships with its own Dockerfile, so it can be deployed on its own.