Alex Beattie

Senior AI/LLM Engineer. Bedrock · Strands · Langfuse · Pinecone · Cohere.

Unified Connectors is an MCP-first, IDE-native workflow layer for engineering context.

I built it because modern software work is spread across too many tools: GitHub for PRs, Jira for tickets, Confluence for decisions, Slack for team context, Google Docs for planning, Miro for diagrams, and Langfuse for model behavior. The actual work happens in the IDE, terminal, Django shell, and AI coding agent.

Unified Connectors brings that context closer to where engineering decisions happen.

What It Does

The core surface is a local MCP server, connectors-mcp, that lets Cursor, Claude Desktop, and MCP-compatible agents query engineering systems directly.

It also exposes:

  • a terminal CLI
  • a Python API
  • a Django management command
  • typed provider actions
  • webhook verification helpers
  • a composite engineering-context brief

The goal is not to replace GitHub, Slack, Jira, Confluence, Google Docs, or Miro. The goal is to create a practical action layer for the repeated workflows that slow down engineering and AI-agent sessions.

Providers

Unified Connectors normalizes actions across:

  • GitHub
  • Slack
  • Jira
  • Confluence
  • Google Docs / Drive
  • Miro

Example workflows:

  • Pull recent PRs and assigned Jira issues before starting a backend task.
  • Search Confluence for architecture context without leaving the IDE.
  • Build a one-shot project context brief across PRs, tickets, docs, and team updates.
  • Post deployment notes or review summaries from scripts.
  • Give an AI coding agent structured project context before it edits code.

Why It Matters For AI Engineering

LLM systems do not just need model context. They need engineering context.

Langfuse can explain how a model behaved. Unified Connectors helps connect that behavior back to the PRs, tickets, specs, docs, and team decisions that explain what changed.

That matters in production AI work because debugging is rarely isolated to one file. A Stella or KiNDD issue might involve:

  • a model route
  • a prompt change
  • a retrieval threshold
  • a frontend rendering issue
  • a Jira requirement
  • a PR discussion
  • a Langfuse trace
  • a Confluence decision

The connector layer is how those pieces become available to the agent and the engineer without opening six browser tabs.

Stack

Language: Python
Protocol: Model Context Protocol
Server: FastMCP
Surfaces: MCP server, CLI, Python API, Django management command
Providers: GitHub, Slack, Jira, Confluence, Google Docs/Drive, Miro
Use case: IDE-native engineering context for AI-assisted development

Lessons

The main lesson is that AI development workflows need first-class internal tooling.

Custom rules and prompts help, but agents become more useful when they can retrieve structured project context from the systems where work actually happens. The best engineering agents do not just read the repo. They understand the surrounding workflow.