Bureau MCP

A personal MCP connection aggregator — like a private Composio — that exposes one unified MCP endpoint and routes tool calls to Zo MCP, Context7, Superproductivity, Google Calendar/Tasks, Notion, Mem0 MCP, and Canva MCP.

This project is currently a scaffold/idea stage. The code does not exist yet. The work is captured as a DeepWorkPlan under .dwp/plans/PLAN_bureau_mcp/.

Quick start for AI/CLI coders

If you are an AI agent or a CLI coder (Cline, ForgeCode, Kilo CLI, etc.), start here:

1. Read the DeepWorkPlan README first
   .dwp/plans/PLAN_bureau_mcp/README.md

2. Pick up the first unchecked task
   The README has a task list with [ ] checkboxes. Always start with the first [ ] task and complete it before moving on.

3. Use the ready-to-use prompts
   .dwp/plans/PLAN_bureau_mcp/PROMPTS.md contains copy-paste prompts for execute, resume, status, and modify.

4. Run validations before marking a task done
   Every task file lists validation commands. Run them and confirm they pass.

5. Update logs and progress
   After each task: mark the task [x] in the README, fill the task file's Completion & Log section, and update PROGRESS.md.

Related MCP server: MCP-GET

Project status

• Plan: PLAN_bureau_mcp

• Location: .dwp/plans/PLAN_bureau_mcp/

• Current status: 0/11 tasks completed

• Workspace: Not a git repo yet.

Task list (copied from the plan)

• Task 1: Inventory and classify all target MCP integrations

• Task 2: Define Bureau MCP architecture and schema normalization contract

• Task 3: Scaffold Bureau MCP repo and core gateway skeleton

• Task 4: Implement core proxy, routing, and schema aggregation

• Task 5: Implement Zo MCP and Context7 adapters

• Task 6: Implement Google Calendar/Tasks and Notion adapters

• Task 7: Implement Superproductivity, Mem0 MCP, and Canva MCP adapters

• Task 8: Add end-to-end tests, Zo.Computer deployment config, and Tailscale service documentation

• Task 9: Security Review

• Task 10: Skills & Agents Discovery

• Task 11: Executive Report

What is Bureau MCP?

The goal is to aggregate all of Shane's MCP tools into a single service so agents do not need individual MCP setups.

Target backends:

1. Zo MCP

2. Context7

3. Superproductivity

4. Google Calendar / Tasks

5. Notion

6. Mem0 MCP

7. Canva MCP

Some backends are expected to be stdio and some HTTP/SSE/Streamable HTTP. The first task is to classify each one before scaffolding.

Deployment target: Zo.Computer, a Linux server, as an internal Tailscale service.

Important rules

• Do not commit secrets. No API keys, OAuth tokens, Zo config, or real credentials in any commit. Use environment variable placeholders only.

• One task at a time. Work through the DeepWorkPlan in order.

• Validate before marking done. Tests, lint, typecheck, and build must pass where applicable.

• No guessing Zo config. If Zo MCP details are not available, document the blocker and ask for them.

For CLI coders

Use these commands to inspect the project:

# View the plan
ls .dwp/plans/PLAN_bureau_mcp/

# Read the plan README
cat .dwp/plans/PLAN_bureau_mcp/README.md

# Read ready-to-use prompts
cat .dwp/plans/PLAN_bureau_mcp/PROMPTS.md

# Check current progress
cat .dwp/plans/PLAN_bureau_mcp/PROGRESS.md

When you resume, use the first [ ] task file:

cat .dwp/plans/PLAN_bureau_mcp/1.task_inventory_classify_mcp_integrations.md

Suggested first action

Start with Task 1: inventory and classify every target MCP. This is research-only and tells us how to scaffold the gateway and adapters.

Files you should know about