Strata MCP Server
Enables LLMs to model, validate, and analyze multi-cloud infrastructure as a typed graph, with tools for IaC import/export, cost estimation, and architecture review.
README
Strata
Live app: strata.mk-dir.com

A registry-driven canvas for modeling cloud infrastructure as a typed graph.
Drag services onto a canvas, configure them with auto-generated forms, and connect
them with typed relationships (contains, routes_to, invokes, peers_with, …)
— with a multi-cloud catalog spanning AWS, Google Cloud and Azure. It can also import
existing Infrastructure-as-Code — CloudFormation (JSON/YAML), Terraform
show -json (AWS/GCP/Azure) and Azure ARM — into the same graph, export the graph
back out as IaC, and ingest live cloud state via provider discovery APIs.
On top of the graph sits a layer of pure analysis engines — validation with one-click autofix, true internet-reachability evaluation, an "Explain & Clean" account review (cost + risk + a safe-cleanup checklist), cross-cloud migration mapping, a change/audit receipt between any two versions, tag tinting/filtering, canvas annotations, and a Diagram-as-Code DSL — all exposed both in the UI and over MCP.
MCP server
An LLM/agent can drive Strata over the Model Context Protocol.
The server (src/mcp/server.ts) is a thin wrapper over the same pure engines the
app uses — registry, validation, IaC import/export, and cost — so an agent can
query and transform a graph exactly as the UI does. No DOM, network, or
credentials.
Run it (stdio transport; uses npx tsx, no build step):
npm run mcp
Then point an MCP client at that command, e.g.:
{
"mcpServers": {
"strata": { "command": "npm", "args": ["run", "mcp"], "cwd": "/path/to/strata" },
},
}
Tools: list_services, get_service, validate_architecture,
suggest_rules, import_iac, export_iac, estimate_cost,
review_account (cost + risk + safe-cleanup report), evaluate_reachability
(true internet-reachability, not topology), map_to_cloud (cross-cloud
equivalence), graph_to_dsl / graph_from_dsl (Diagram-as-Code round-trip),
list_autofixes / apply_autofix (deterministic validation fixes),
change_receipt (audit receipt of what changed between two graphs),
tag_report.
Live cloud discovery is separate — it runs through a Cloud Control SDK route (
src/app/api/discover/route.ts);src/aws/mcp.tsis the pure transform behind it. See CONTRIBUTING.md for the architecture.
The core idea: everything visual is derived from a data registry, not hardcoded. Adding a new AWS service is a single catalog entry — no UI changes.
Strata is open source. Contributions welcome.
Documentation
Strata ships its docs as part of the app, served at /docs, in two public
sections:
- User Guide (
/docs/guide) — how to use the app: building diagrams, importing IaC, validation, and saving/loading. - Architecture & Engineering (
/docs/architecture) — how it works internally: the service registry, domain model, persistence, MCP/IaC import, testing, and the roadmap.
The docs are authored as Nextra MDX under src/content/.
Quick Start
Strata is now a single app: npm run dev serves the product at / and the docs
at /docs.
npm install
npm run dev
- Product: http://localhost:3000/
- Docs: http://localhost:3000/docs
Saved diagrams persist in your browser (localStorage) — no external
infrastructure required, and it works on a read-only serverless host. Use
Export / Import JSON to move a diagram between browsers or share it. (The
server-side Repository + /api/graphs route remain for a future durable
backend, but the app no longer relies on them for saving.)
Other scripts
npm run build # production build (compiles the product and prerenders the docs)
npm run start # run the production build
npm run lint # lint
npm test # run the Vitest suite
Configuration
| Env var | Default | Purpose |
|---|---|---|
AWS_FLOW_REPOSITORY |
file |
Backend for the retained server tier (not used by browser-local save). |
AWS_FLOW_DATA_DIR |
.data/graphs |
Directory for the file-backed store (retained server tier only). |
AWS_FLOW_API_TOKEN |
(unset) | If set, the graph API requires Authorization: Bearer <token>. |
NEXT_PUBLIC_STRATA_HOSTED |
(unset) | Set to 1 on any shared/hosted deployment (see below). |
STRATA_INTEREST_WEBHOOK |
(unset) | Incoming-webhook URL to notify on "coming soon" interest (see below). |
"Coming soon" interest notifications
When NEXT_PUBLIC_STRATA_HOSTED=1, clicking a "coming soon" prompt's "I'd
use this" button POSTs a tiny { feature, note? } signal to the same-origin
/api/interest route (it's skipped locally — there's nothing to collect on a
single-user box). That route, with no datastore and no npm dependency:
- always logs the signal (so it shows in your host's function logs, e.g. Vercel), and
- if
STRATA_INTEREST_WEBHOOKis set, forwards a short message to that incoming-webhook URL. Use whatever gives you a notification with zero infra — a Slack or Discord incoming webhook (phone push), ntfy.sh (push), or a webhook→email relay. The body sends bothtext(Slack) andcontent(Discord), so one URL works for either. No credentials are involved, and the endpoint accepts only a small validated payload.
Live discovery & credentials
The Connect to AWS → Live scan flow runs server-side. With no credentials in the request, it uses the server process's default credential chain (env vars / shared profile / SSO / instance role) — fine for a single-user local run, where those are your own credentials.
On a shared/hosted deployment that ambient chain would be the operator's
account, so any visitor could enumerate it. Set NEXT_PUBLIC_STRATA_HOSTED=1
(at build and runtime) to disable the ambient fallback: each user must then
bring their own AWS credentials, entered in the modal and sent over HTTPS for a
single scan. Those keys are used in-memory only — never written to disk, logged,
returned, or saved into a diagram. Users should supply temporary, read-only
credentials (e.g. aws sts get-session-token, or an assumed ReadOnlyAccess
role). The credential-free Paste export tab remains available either way.
Project Structure
src/
aws/ Registry + domain model (the data foundation)
types.ts ServiceDefinition, ConfigField, RelationshipKind, scopes
registry.ts Aggregates catalogs; lookups, search, validateRegistry()
categories.ts Category + relationship presentation metadata
model.ts InfrastructureGraph / ResourceInstance / Relationship, validateGraph()
regions.ts AWS region reference list
rules.ts Architecture validation + best-practice rule suggestions
relationshipClasses.ts Edge visual encoding (colour/dash/arrowhead per class)
overlays.ts Analytical overlays (IAM-trust, network path, heat, reachability, tags) — pure
reachability.ts Policy-aware internet-reachability eval (subnets/routes/SG ports) — pure
review.ts "Explain & Clean" account review: cost + risk + safe-cleanup checklist — pure
cloudMap.ts Cross-cloud equivalence / migration (serviceId+provider rewrite) — pure
autofix.ts Deterministic fixes for validation findings (detect + apply) — pure
receipt.ts Change/audit receipt between two graphs (drift + cost + findings) — pure
tags.ts Tag collection + tint map + tag filter — pure
annotations.ts Notes / callouts / zones pinned to the canvas (excluded from rules/cost/IaC) — pure
dsl.ts Diagram-as-Code: graph ⇄ YAML round-trip — pure
iac.ts IaC import (CloudFormation + Terraform → InfrastructureGraph)
iacExport.ts IaC export (InfrastructureGraph → CloudFormation / Terraform scaffold)
mcp.ts Pure transform: DiscoveredResource[] → graph (mapDiscoveredToGraph)
discovery.ts Cloud Control descriptions → DiscoveredResource[] (no AWS SDK here)
services/*.ts Per-category service catalogs (networking.ts is the template)
canvas/ Pure geometry + containment layout (geometry.ts, layout.ts)
server/ Retained server tier (kept for a future durable backend)
repository.ts Repository interface
fileRepository.ts Default file-backed store
graphSchema.ts Zod schema for graph validation
auth.ts Optional bearer-token guard (AWS_FLOW_API_TOKEN)
index.ts getRepository() — backend selection via env
lib/
localStore.ts Browser localStorage save/load (the active persistence path)
api.ts Fetch client for the graph + discover APIs (discover is live)
app/
(product)/ The Strata canvas app, served at /
(docs)/ Nextra docs, served at /docs
api/graphs/ Graph REST Route Handlers (GET/POST, GET/PUT/DELETE by id)
api/discover/ Live discovery Route Handler (Cloud Control SDK, server-only)
components/ UI: Palette, Canvas, Inspector, CommandPalette
hooks/ Canvas state, rendering, interaction, undo/redo
content/ Nextra MDX docs (User Guide + Architecture), served at /docs
Adding a new AWS service
Append one ServiceDefinition to the matching catalog in src/aws/services/
(use networking.ts as the template) — the palette, colours, icons, inspector
form, and search pick it up automatically. See the
Service Registry docs.
Architecture
For the full design — registry schema, domain model, the registry-driven UI, the
swappable persistence layer, MCP ingestion readiness, and a candid list of gaps and
next steps — see the Architecture & Engineering docs
(src/content/architecture/).
Roadmap
What's shipped, the active follow-ups, and what's deferred until a backend exists: see ROADMAP.md.
Tech Stack
Next.js (App Router) · React · TypeScript · Tailwind CSS · Nextra (docs).
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.