🔌 MCP Server Demo

Live Model Context Protocol server running on this infrastructure. Agentmemory MCP with semantic search, memory consolidation, and 8 tool endpoints.

agentmemory MCP
Checking...

🧠 Memory Recall

Semantic + keyword search across all stored memories. Powers context-aware AI agents.

Results appear here...

📋 Available MCP Tools

8 tools registered through the agentmemory MCP server:

  • memory_recall — search past observations
  • memory_save — persist insights
  • memory_smart_search — hybrid semantic search
  • memory_sessions — browse session history
  • memory_consolidate — tiered memory pipeline
  • memory_diagnose — health checks
  • memory_lesson_save — save learnings
  • memory_reflect — knowledge graph traversal

📊 Server Stats

Status —
Functions 257
Port 3111
Integration Hermes Agent + MCP

🤔 Why MCP?

Before MCP, connecting AI agents to tools meant writing custom integrations for every framework — a separate adapter for LangChain, another for Claude SDK, another for OpenAI. Every tool needed n implementations.

MCP standardizes this. It's like USB-C for AI tools: one protocol, any agent, any tool. Build a tool server once, any MCP-compatible agent can use it — the Hermes Agent, Claude Desktop, Cursor, or any custom agent implementing the protocol.

The demo below uses an agentmemory MCP server — search, save, consolidate, reflect — all accessible through a single protocol interface.

💰 Cost & Architecture

🖥️
Edge Function
Runs on Cloudflare Pages — 35+ locations, 0 servers to manage
🔌
ZERO Backend Calls
Search is purely local BM25 matching on 25 seeded memories
⚡
ZERO API Tokens
No LLM calls, no embedding generation, no smart search overhead
$0
Monthly Cost
Cloudflare Pages free tier — no compute, no database, no egress fees

✅ Advantages Over Alternatives

Approach Integration Cost Security
MCP (this demo) Universal — any MCP agent $0 / no tokens Self-contained edge function
REST API proxy Custom per framework Compute + bandwidth Exposes backend directly
Embedding-based RAG SDK-dependent ~$0.01/query (embeddings) API key management
Direct LLM memory Single-agent only Context window cost Data in prompt history

❓ FAQ

How is this different from a regular REST API?

REST APIs require a separate SDK or client for every agent framework. MCP is a standard protocol that any MCP-compatible agent can discover and call dynamically — including tool descriptions, input schemas, and error handling — without custom glue code.

Does this consume my API quota?

No. The search functionality uses local keyword matching on the edge — zero LLM calls, zero embedding calls, zero API tokens. The health check returns hardcoded data. This demo costs exactly $0 to run.

Can I use this in production?

This demo is a proof of concept. For production, you'd deploy a real MCP server with your own data, proper authentication, and a backend database. The architecture pattern — edge function proxying to MCP — is production-ready. Contact me for a production deployment.

What agents support MCP?

Claude Desktop, Cursor, Windsurf, the Hermes Agent, and custom agents using the MCP SDK (Python, TypeScript, Kotlin, Java). The protocol is open-source and growing fast — over 1,000 servers in the registry as of mid-2026.

What's the difference between MCP and function calling?

Function calling is built into individual LLM providers (OpenAI, Anthropic). MCP sits between — it's a standard for exposing tools that any provider's function-calling layer can consume. Your tool server works across providers without rewrites.

How do I build my own MCP server?

The MCP TypeScript SDK has a McpServer class — define tools with schemas, expose resources, register prompts. The Hermes Agent blog covers this in depth. Or I can build one for your stack.

Want an MCP Server for Your Stack?

I build custom MCP servers for Ghost CMS, CRMs, knowledge bases, chatbots, and any API — with auth, rate limiting, and monitoring included.

View MCP Services →