This MCP server gives an AI assistant accurate, source-backed answers about
Smithy. It retrieves
semantic matches from the official Smithy documentation through an Amazon Bedrock
Knowledge Base, and exposes the published Markdown directly for browsing and full-document
reads. Point any MCP client at one HTTPS endpoint and start asking.
POSThttps://smithymcp.com/mcp
Stateless MCP Streamable HTTP. Send JSON-RPC 2.0 requests to this URL.
Add to your assistant
Pick your tool. Kiro supports a real one-click install; Claude Code and Codex install with
a single terminal command. The raw JSON / TOML config is shown alongside each as a
fallback.
Kiro
One-click
Adds the server to your Kiro user MCP config. Kiro asks you to approve it first.
+ Add to Kiro
Or add it manually (.kiro/settings/mcp.json)
This server answers questions about Smithy. For live feedback
while you write a .smithy model — diagnostics, hover,
completion, go-to-definition — run the official
Smithy Language Server
(LSP) in your editor. The two work well together: the LSP keeps your model valid as you
type, and this MCP server explains the concepts behind the rules.
1. Get the language server
Install the Smithy extension for VS Code (it bundles and launches the
language server for you), or run the server directly for another editor:
# Run the Smithy Language Server (Java 17+ on PATH)
# Coursier is the simplest launcher:
cs launch software.amazon.smithy:smithy-language-server:latest.release
The VS Code Marketplace extension is the easiest path and needs no manual launch; the
command above is for editors that speak LSP directly (Neovim, Emacs, Zed, …).
2. Point your editor at it
Most editors auto-detect a Smithy project from a smithy-build.json at the
workspace root. A minimal one:
{
"version": "1.0",
"sources": ["model"]
}
Then a model file the language server will check as you edit it:
Save it and the language server reports unknown shapes, missing traits, and syntax errors
inline. When it flags something you do not understand, ask your assistant — it calls
this MCP server to explain the rule.
Together: the LSP tells you that line 12 is wrong; the Smithy
MCP server tells you why, with the doc passage that explains the trait.
Example prompts
Once configured, ask your assistant things like:
“Search the Smithy docs for how to define a service.”
“How do resource shapes and their lifecycle operations work in Smithy?”
“My model has @pattern on an integer — why is the LSP
complaining?”
“Show me an example of applying a custom trait.”
Architecture
AWS Lambda — serverless compute for the MCP server.
Bedrock Knowledge Base — semantic search over S3 Vectors storage.
CloudFront + WAF — global CDN, TLS, and per-IP rate limiting.
S3 — documentation storage and vector embeddings.
Documentation is refreshed from smithy-lang/smithy weekly, then re-indexed.