Skip to main content
Background

Deploy a serverless MCP server with WebAssembly

Learn to build an LLM chatbot with Akamai Functions and GPU-powered inference, then deploy it globally.

Timestamped summary

Timestamped summary

[00:00–01:01] Introduction: Thorsten Hans introduces the concept of building a serverless Model Context Protocol (MCP) server, outlining the video structure which covers the decision-tracking MCP server example, building blocks used, code implementation, deployment, and client configuration.

[01:01–02:53] Use case context: Overview of the decision-tracking MCP server, built to track and retrieve software architecture decisions (such as OAuth 2.0 and OIDC authentication) directly within developer editors and IDEs using chat interfaces.

[02:53–05:18] Building blocks overview and wasmcp: Explanation of the three main components: wasmcp (an open source framework/toolset for building MCP servers with WebAssembly components), Spin, and Akamai Functions, including a look at the wasmcp GitHub repository.

[05:18–07:42] Building a Hello World MCP server: Demonstration of creating a new Rust-based MCP server using wasmcp new hello-wasmcp --language rust, inspecting lib.rs for add and subtract tools, compiling to Wasm, registering it, and composing mcp_server.wasm.

[07:42–09:58] Spin framework and local testing: Explanation of Spin as a CNCF developer tool and runtime, followed by running spin up -f mcp_server.wasm to host the MCP server locally on [http://127.0.0.1:3000](http://127.0.0.1:3000).

[09:58–11:03] Testing locally in VS code: Connecting the local MCP server URL to VS Code, confirming active tools (add and subtract), and executing an addition tool call via the AI chat agent.

[11:03–12:29] Akamai Functions overview: Overview of Akamai Functions as a globally distributed, multitenant serverless platform that automatically routes requests to the nearest region for low latency without requiring manual infrastructure management.

[12:29–14:09] Deploying Hello World to Akamai Functions: Configuring spin.toml to reference mcp_server.wasm, deploying via spin aka deploy, and testing the deployed remote server URL in VS Code with a sample calculation tool call.

[14:09–16:56] Decision log MCP server implementation: Walkthrough of the decision log MCP server, inspecting spin.toml (enabling key_value_stores = ["default"]) and decision.rs for tool definitions (list_decisions, get_decision_by_slug, insert_decision, delete_decision) and key-value store interactions using serde and the Spin SDK.

[16:56–18:07] Deploying and testing decision log server: Composing, deploying the decision-log MCP server to Akamai Functions, adding its live endpoint into VS Code, and executing commands to list and track decisions in the chat agent.

[18:07–20:15] Conclusion: Summary of the benefits of serverless Wasm deployment on Akamai Functions (submillisecond start-up times, no server management).

Share