Production MCP Server: 365+ Tools
A production MCP server deployed on Cloudflare Workers that exposes 365+ tools across 23 platform integrations. Every AI agent, content pipeline, and automation in the Trending Society platform flows through this server, from article generation to social media publishing to newsletter delivery.
Why An MCP Server
AI agents are only as useful as the tools they can access. Instead of hardcoding API calls into every agent, this MCP server acts as a unified tool layer. Any agent that speaks the Model Context Protocol can discover and invoke 365+ tools without knowing the underlying API details.
Architecture
The server runs on Cloudflare Workers at the edge, with KV for session storage. Each integration is a self-contained module that registers its tools at startup:
- Tool registration: each integration declares its tools with typed input schemas
- Session management: KV stores per-session state, enabling multi-turn agent conversations
- Protocol compliance: supports both Streamable HTTP (modern) and SSE (legacy) transports
- Auth routing: each integration authenticates independently. Supabase uses service role keys, GitHub uses PATs, Langfuse uses public/secret key pairs
Platform Integrations
| Platform | Tools | Capabilities |
|---|---|---|
| Google Workspace | 64 | Gmail, Calendar, Drive, Sheets, Docs, Slides, Tasks, Forms, Meet |
| Linear | 60 | Issues, projects, cycles, teams, labels, users |
| Stripe | 34 | Customers, products, prices, subscriptions, invoices, payments |
| Resend | 29 | Emails, domains, API keys, audiences, contacts, broadcasts |
| Webflow | 21 | Sites, pages, collections, CMS items, assets, custom code |
| ElevenLabs | 20 | Text-to-speech, voice cloning, dubbing, sound effects, audio isolation |
| Shopify Admin | 19 | Products, orders, customers, inventory |
| Content Library | 16 | Source ingestion, AI analysis, affiliate matching, brand kit |
| Production Studio | 15 | Presets, templates, shots, recipes, video composition |
| Shopify Storefront | 14 | Headless commerce: cart, checkout, search |
| OpenAI | 12 | Chat, embeddings, images, audio, moderation, files, fine-tuning |
| Twilio | 11 | SMS, MMS, voice calls, recordings |
| Airtable | 10 | Bases, records, batch operations, comments |
| Cloudinary | 8 | Media upload, transformation, video processing, asset management |
| Anthropic | 7 | Messages, token counting, batches |
| GitHub | 5 | Repository audit, commit history, PR management, CI status |
| Langfuse | 5 | Prompt registry, trace inspection, cost analysis |
| Tavily | 4 | AI-powered web search and content extraction |
| Replicate | 3 | ML model inference, prediction tracking, cancellation |
| Supabase | 3 | Database queries, schema inspection |
| Telegram | 3 | Bot messaging, webhook management |
| Apify | 2 | Structured data extraction, dataset retrieval |
| Inoreader | 2 | RSS feed automation, article ingestion |
Engineering Decisions
Cloudflare Workers over a traditional server: sub-50ms cold starts, global edge deployment, and KV provides session persistence without a database. The server handles bursty agent traffic without provisioning concerns.
Tool groups over individual tools: the MCP spec allows grouping related tools under a single registration. Instead of exposing 365 individual endpoints, tools are organized into logical groups by platform, reducing discovery overhead for agents.
Typed schemas everywhere: every tool input is validated against a Zod schema before execution. Invalid agent requests fail fast with actionable error messages instead of silently producing bad data.
Deployment
The server deploys via GitHub Actions on push to main. Path filters ensure only changes to servers/mcp/** trigger a rebuild. Wrangler handles the Cloudflare deployment, and the server is live at mcp.trendingsociety.com.
By The Numbers
- 365+ tools across 23 platform integrations
- Sub-50ms cold start on Cloudflare Workers edge
- 100+ articles/week processed through the pipeline tools
- Full MCP 2025-11-25 spec compliance with Streamable HTTP + SSE
