activity-mcp 👨‍🏭

Suite of tools for interacting with a number of services, like Slack, Harvest, GitHub, etc.

Fully automate:

• Timesheets
• Retro notes
• Project tickets
• Async standups
• General admin tasks, checking emails, responding to events
• Deep analysis, summarization and search of any service

Built to eliminate purely transformational tasks, recollection work, and administrative overhead.

Why does this exist?

AI was always supposed to save us from tedium. That was the promise.

To bring us to a gilded age where people would only do work that interested them.

Art, programming, music, creative writing, problem solving.

What we see instead is commercial AI taking the human's creative space, leaving us humans to toil solely in project management tools, meetings, and emails.

MCP fits nicely into this by connecting AI agents, who can contain any context, to external services. But not just input, MCP allows AI agents to read from external services.

However, the best MCP tooling is unwieldy and difficult to use for anyone who is not a developer.

This means that all-in-one tools like Copilot, Perplexity (somewhat), and friends end up topping charts.

Despite not being as personalised or as powerful as other tools, they are simple to use and setup.

It should be possible for non-technical, non-poweruser individuals to see the benefits that personalised MCP tools can bring to their workflow and sanity.

In fact, it should be easy.

That is the point of this project.

Project Structure

activity-mcp/
└── packages/
    ├── mcp-server/          # Core MCP server with service integrations
    ├── desktop-app/         # Electron wrapper application
    ├── client-workflows/    # Semi-automated workflow scripts
    └── auth-middleware/     # OAuth authentication middleware

Installation

Use a version manager like fnm, nvm, Nix, etc to install the .node-version:

fnm use

Enable pnpm with corepack and install the dependencies:

corepack enable

pnpm install

Environment variables

Tools throw errors if required environment variables are missing.

Required:

• OPENAI_API_KEY - OpenAI API key for AI functionality

Service Integrations:

• SLACK_USER_TOKEN - User token with search scopes (requires Slack app)
• SLACK_BOT_TOKEN - Bot token for posting messages
• SLACK_CLIENT_ID - OAuth client ID from Slack app
• SLACK_CLIENT_SECRET - OAuth client secret from Slack app
• SLACK_STATE_SECRET - Any random string for OAuth state validation
• CLICKUP_TOKEN - ClickUp API token
• CLICKUP_TEAM_ID - ClickUp team ID from any URL
• HARVEST_ACCESS_TOKEN - Harvest API token
• HARVEST_ACCOUNT_ID - Harvest account ID
• GITHUB_TOKEN - GitHub personal access token

File Paths:

• GIT_REPOSITORIES_ROOT_DIRECTORY - Root folder containing Git repos
• TIME_ENTRIES_SQL_PATH - Path to Timing.app SQLite database
• GOOGLE_TOKEN_PATH - Path to Google OAuth token.json
• GOOGLE_CREDENTIALS_PATH - Path to Google OAuth credentials.json

Google Calendar Setup

1. Create project in Google Cloud Console
2. Enable Calendar API
3. Create OAuth 2.0 Desktop credentials
4. Download credentials.json to project root
5. Run app once to generate token.json

Slack App Setup

1. Create app at api.slack.com/apps
2. Add OAuth scopes: chat:write, channels:read, app_mentions:read
3. Install to workspace to get tokens
4. Copy Client ID and Secret from Basic Information

Running locally

As of now, there isn't a unified way to run the whole stack at once.

Each package requires independant setup and has a unique way to run.

desktop-app is the simplest to run.