Metawork.Studio · Brain

What was built

Metawork Brain is the internal operating system we use to run our agency. It is built as an MCP server, which means Claude and every other AI assistant we use can query our knowledge base, our project tracker and our pipeline without anyone pasting context into a chat window.

Everything lives in one workspace backed by Git. Client pipelines, meeting transcripts, make.com scenario documentation, onboarding notes, playbooks. Full version history, clean diffs, no vendor lock in.

Knowledge flows in from everywhere

Brain pulls from project management tool for tasks and timelines, Fathom for meeting transcripts, Cap for video explainers, Pipedrive for deal and meeting data, and our own Make scenarios for automation state. Context is created along the way in Claude and ChatGPT sessions, organized as compiled md files and as a graph in our project management tool. All of it syncs into structured Markdown under a Projects folder, one directory per client.

The best part

Due to detailed context tracking, switching AI app is a matter of connecting the MCP in ChatGPT instead of Claude. ChatGPT immediately understands what needs to be done, where to look up information and how to work. No provider lock in. No memory loss. Connect to MCP, continue the work.

What the team can do in chat

The MCP layer is the core of the system. Ask Claude what happened in the last check in with a client and it reads the transcript. Ask it to create a new project and it scaffolds the folder structure, the YouTrack project and the starting docs in one go. The full toolset covers:

• Creating, updating and linking project issues, including subtask linking which essentially provides the graph layer of the context
• Automated time tracking, so hours land in the right category
• Project creation with filesystem scaffolding
• Reading meeting history, materials and client documentation
• Workload planning across the team
• Generation of branded documents: proposals, specs, documentation, status reports, handoffs

The /workload view packs estimates into a realistic schedule per person. A Friday cron emails every teammate a link to their own workload page, so Monday planning is already done before Monday.

Several other server-side tools cover website updates whenever a project is finished, linking of emails into deals and projects inside our Pipedrive CRM, sync of milestones between the CRM and project management tool. UI layer provides a convenient way to edit website content, as well as handling of client portals.

The client portal

Brain also serves the outside world. Every client gets a portal with a milestone board driven straight from project management tool, a timeline of past meetings and materials fed from the filesystem, and a questions und feedback section where questions, requests and bugs become tickets for the team. Everything in real time. An "Im Kalender" block reads upcoming meetings from Pipedrive per project and updates itself.

Portals run in German or English per client.

Tech stack

• MCP server exposing Brain to Claude and other assistants
• YouTrack for issues, milestones, time tracking and the portal milestone board
• Pipedrive for deals and calendar data
• Fathom for meeting transcripts
• Cap for video explainers
• Make for the automation scenarios Brain documents and monitors
• Git as the storage backend, Markdown as the format
• Cloudflare Access with multiple identity providers for team auth
• Webhook driven restarts so tool changes go live without manual intervention