MCP: The Protocol That Makes AI Agents Useful

Model Context Protocol (MCP) is what turns a chatbot into an agent. Without it, AI models can only work with what's in their context window. With it, they can reach into your actual tools and systems.

What MCP Does

MCP provides a standardized way for AI models to:

• Read from external systems (Jira tickets, Confluence pages, Slack messages)
• Write to external systems (create issues, post comments, update pages)
• Search across systems (JQL queries, CQL queries, full-text search)

My MCP Stack

Practical Patterns

Pattern 1: Cross-System Correlation

Agent reads Slack channel → extracts ticket references
Agent queries Jira for those tickets → gets status
Agent reads Confluence for sprint goals → cross-references
Agent generates digest with risk flags

Pattern 2: Write-Back with Approval

Agent parses standup notes → extracts per-ticket updates
Agent previews proposed Jira comments → waits for approval
User approves → Agent posts comments to Jira

Pattern 3: Enrichment Pipeline

Calendar event → Jira context → Confluence docs → Slack threads
= Meeting prep with full project context

Gotchas

1. Auth management — MCP servers need tokens. Glean tokens expire hourly. Plan for this.
2. Rate limits — Jira API has rate limits. Batch your queries.
3. Data volume — Slack channels can return massive payloads. Use targeted queries.
4. Error handling — MCP calls can fail silently. Always validate responses.