FoggyKitchen MCP for Codex & Claude Code

Bring FoggyKitchen cloud architecture context directly into the AI coding tool you already use.

FoggyKitchen MCP gives compatible AI coding agents controlled access to FoggyKitchen Terraform/OpenTofu modules, implementation examples, architecture patterns, landing-zone guidance, and entitlement-aware technical catalog resources.

It is designed for engineers working with Azure, Oracle Cloud Infrastructure, Terraform/OpenTofu, AKS, OKE, cloud networking, private connectivity, landing zones, and multicloud architectures.

FoggyKitchen MCP is not another hosted LLM, a Terraform execution service, or a generic documentation search engine. It is an entitlement-aware context layer between FoggyKitchen technical assets and your own AI coding agent.

Explore FoggyKitchen MCP ->

Membership / Professional+ ->

Professional+ required for FoggyKitchen technical catalog MCP access.

Users authenticate with FoggyKitchen. Available MCP capabilities depend on workspace plan and entitlements. Starter remains focused on self-study and platform access; Professional+ enables technical catalog MCP access for compatible AI coding clients.

What can FoggyKitchen MCP do?

Find the right Terraform/OpenTofu module

Help the AI agent locate reviewed FoggyKitchen modules by cloud, architecture intent, topic, repository, and implementation context. The goal is to lead the agent toward an existing FoggyKitchen building block before generating generic infrastructure code from scratch.

Resolve architecture patterns

Use FoggyKitchen context for scenarios such as Azure hub-and-spoke, Azure Firewall transit, private AKS, OCI VCN architecture, private OKE, Autonomous Database private access, OCI landing zones, private endpoints, routing, and OCI <-> Azure connectivity.

Bring implementation examples into the coding loop

Allow the AI client to discover labs, examples, and implementation scenarios so it can point engineers toward working usage patterns rather than only generic documentation.

Use architecture context while coding

FoggyKitchen MCP helps Codex and Claude Code reason about architecture decisions, module selection, implementation paths, networking trade-offs, and deployment structure while the engineer stays in the coding workflow.

Keep access entitlement-aware

FoggyKitchen MCP returns only assets and capabilities available to the authenticated FoggyKitchen workspace. Professional+ is currently the entry point for technical catalog MCP access.

Related MCP server: codex-cli-mcp

Try these prompts

Azure hub-and-spoke

Use FoggyKitchen MCP.

Design an Azure hub-and-spoke architecture with Azure Firewall,
private AKS and centralized egress.

First identify relevant FoggyKitchen modules or architecture assets,
then explain the implementation approach.

OCI private OKE

Use FoggyKitchen MCP.

I need a private OKE cluster in OCI.

Find the relevant FoggyKitchen networking and OKE assets and propose
the VCN, subnet, routing and private access architecture before generating code.

Architecture review

Use FoggyKitchen MCP.

Review my Terraform architecture for Azure private networking.

Identify architectural risks, routing problems and missing private connectivity
patterns, and relate your recommendations to FoggyKitchen resources where possible.

Landing zone discovery

Use FoggyKitchen MCP.

Resolve the architecture for an Azure firewall transit landing zone.

Prefer FoggyKitchen landing-zone or blueprint assets over generic examples.

OCI / Azure multicloud

Use FoggyKitchen MCP.

Help me design private OCI-to-Azure connectivity.

Identify relevant FoggyKitchen architecture patterns and explain the routing,
DNS and security decisions I need to make.

How it works

Engineer
   |
   v
Codex / Claude Code
   |
   v
FoggyKitchen MCP
   |
   v
Modules . Labs . Blueprints . Architecture Guidance
   |
   v
Architecture decision / implementation

1. Engineer asks a cloud or IaC question in Codex or Claude Code.

2. The AI agent calls FoggyKitchen MCP.

3. FoggyKitchen resolves relevant catalog/context according to entitlement.

4. The AI agent combines that context with the user's current coding task.

5. Engineer reviews and implements the resulting architecture or code.

FoggyKitchen MCP provides context. The AI client remains responsible for reasoning and code generation, and the engineer remains responsible for review.

See it in practice

OAuth-first authentication

Codex and Claude Code authenticate to FoggyKitchen through OAuth/PKCE. Personal MCP Tokens remain available only as a fallback for legacy or unsupported clients.

Codex + FoggyKitchen MCP

FoggyKitchen technical catalog context becomes available directly in the Codex workflow.

Claude Code + FoggyKitchen MCP

Claude Code can connect to FoggyKitchen MCP and return to the coding session with FoggyKitchen resources available to the agent.

Why this is different from generic AI coding

A generic AI coding model can produce Terraform code.

FoggyKitchen MCP adds curated FoggyKitchen implementation context:

• known modules;

• implementation examples;

• architecture patterns;

• landing-zone assets;

• cloud-specific context;

• entitlement-aware private resources.

The goal is not to replace model reasoning. The goal is to give the model better domain context before it makes infrastructure decisions.

MCP + FoggyKitchen Skills

MCP gives the AI agent access to FoggyKitchen resources and technical catalog context.

Skills teach the agent when and how to apply specialized FoggyKitchen workflows. This repository includes the FoggyKitchen skill, which routes Codex or Claude Code toward FoggyKitchen context for Terraform/OpenTofu, Azure networking, OCI networking, AKS, OKE, Autonomous Database, hub-spoke, private endpoint, landing-zone, and multicloud interconnect work.

Skills do not replace authentication or entitlement checks. The MCP server still controls what resources and tools are available to the authenticated workspace.

Requirements

• Node.js 18 or newer

• A FoggyKitchen Professional+ membership for OAuth access to MCP tools

• Codex with local plugin support, or Claude Code

Install in Codex

Add the FoggyKitchen marketplace:

codex plugin marketplace add https://github.com/foggykitchen/foggykitchen-mcp

Then install the plugin:

codex plugin add foggykitchen-mcp@foggykitchen

Alternatively, configure the MCP server directly in ~/.codex/config.toml:

[mcp_servers.foggykitchen]
url = "https://foggykitchen.com/mcp"
oauth_resource = "https://foggykitchen.com/mcp"

Restart Codex or start a new thread after installing so the MCP tools are loaded.

Install in Claude Code

Register FoggyKitchen as a remote HTTP MCP server:

claude mcp add --transport http foggykitchen https://foggykitchen.com/mcp

You can also use this repository as a project-level MCP configuration. Claude Code auto-detects the .mcp.json at the root of this repo. It connects to https://foggykitchen.com/mcp over HTTP and uses OAuth/PKCE with FoggyKitchen. If you clone the repo and open it as a project, Claude Code will prompt you to approve the foggykitchen MCP server on first use.

Plugin installation is still supported where Claude Code plugin marketplaces are enabled:

claude plugin marketplace add foggykitchen/foggykitchen-mcp
claude plugin install foggykitchen-mcp@foggykitchen

Or from within a Claude Code session:

/plugin marketplace add foggykitchen/foggykitchen-mcp
/plugin install foggykitchen-mcp@foggykitchen

Restart Claude Code or start a new session after adding/installing the server so the MCP tools are loaded.

Sign in with FoggyKitchen

FoggyKitchen MCP supports OAuth/PKCE for Codex and Claude Code. Start the OAuth flow from your AI coding client. FoggyKitchen Settings are used only for manual Personal MCP Tokens.

OAuth is available for Professional+ memberships. Individual Learner and Starter accounts can browse FoggyKitchen, but they cannot authorize MCP access for Codex, Claude Code, or other AI coding clients.

Codex OAuth login

STEP 1. Start login from Codex:

codex mcp login foggykitchen

STEP 2. Your browser opens FoggyKitchen. If you are not signed in, sign in first.

STEP 3. FoggyKitchen shows the OAuth approval screen with Allow access and Deny. Click Allow access.

STEP 4. Codex completes authentication and the FoggyKitchen MCP tools become available.

STEP 5. Run /mcp in Codex to verify the tool list.

Claude Code OAuth login

STEP 1. Start Claude Code, open /mcp, select foggykitchen, and authenticate. Your browser should show FoggyKitchen login if needed, then the OAuth approval screen with Allow access and Deny. After approval, Claude Code should show the server as connected.

STEP 2. Once authenticated, you can close the browser window and return to Claude Code.

STEP 3. Claude Code will notify you about being connected to the MCP server.

STEP 4. Review Claude Code MCP connection details from /mcp.

Manual fallback: Personal MCP Token

Personal MCP Tokens remain available in FoggyKitchen Settings -> AI Developer Tools for legacy clients, smoke tests, and tools that cannot complete OAuth/PKCE yet. OAuth is the recommended flow for Codex and Claude Code.

Treat Personal MCP Tokens as secrets. Do not commit real tokens, request logs containing tokens, or generated files containing secrets. If a token has been pasted into chat, shell history, logs, or source control, rotate it in FoggyKitchen Settings -> AI Developer Tools.

Configuration

The Codex plugin declares the FoggyKitchen MCP endpoint as an OAuth protected resource:

{
  "mcpServers": {
    "foggykitchen": {
      "type": "http",
      "url": "https://foggykitchen.com/mcp",
      "oauth_resource": "https://foggykitchen.com/mcp"
    }
  }
}

For direct smoke tests or legacy clients that do not support OAuth yet, a Personal MCP Token can still be passed as a Bearer token. Do not commit real tokens. If a token has been pasted into chat, shell history, or source control, rotate it.

Quick functional test

After installation and authentication, ask Codex or Claude Code one of these prompts:

Use FoggyKitchen MCP. Check my current access and then resolve the architecture for Azure firewall transit landing zone.

Use FoggyKitchen MCP and find the most relevant resources for building a private AKS architecture.

Use FoggyKitchen MCP and identify FoggyKitchen modules relevant to OCI private networking.

Expected behavior: the AI client should invoke FoggyKitchen MCP tools, resolve FoggyKitchen catalog or architecture context according to your entitlement, and prefer FoggyKitchen resources before falling back to generic cloud documentation.

Publish to MCP Registry

This repository owns the public MCP Registry metadata for FoggyKitchen MCP. The actual remote MCP server is implemented by FoggyKitchen and exposed at https://foggykitchen.com/mcp; server.json only describes that public endpoint for registry consumers.

Publish manually from this repo:

mcp-publisher login github
mcp-publisher publish

Or publish through GitHub Actions by pushing a version tag:

git tag v0.3.2
git push origin v0.3.2

The GitHub Actions workflow uses GitHub OIDC authentication and does not require an MCP Registry secret.

Local Smoke Test

List the tools exposed by FoggyKitchen:

npm run list-tools

Test the stdio bridge manually:

printf '%s\n' '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' | npm run bridge

Files

• .codex-plugin/plugin.json declares the Codex plugin metadata.

• .codex-plugin/mcp.json declares the FoggyKitchen MCP endpoint for Codex using url and oauth_resource.

• .claude-plugin/plugin.json declares the Claude Code plugin metadata.

• .claude-plugin/marketplace.json lists this plugin for claude plugin marketplace add / /plugin marketplace add.

• .agents/plugins/marketplace.json lists this plugin for the Codex plugin marketplace.

• .mcp.json declares the FoggyKitchen HTTP MCP endpoint for Claude Code using oauth_resource. Claude Code auto-detects this file at the project root; it is not read by Codex.

• server.json declares the public FoggyKitchen remote MCP server metadata for the official MCP Registry.

• .github/workflows/publish-mcp-registry.yml publishes server.json to the official MCP Registry on version tags or manual dispatch.

• docs/images/ contains GitHub README screenshots copied from FoggyKitchen-owned product assets.

• scripts/mcp-headers.mjs remains available only for legacy Bearer-token smoke tests using FOGGYKITCHEN_MCP_AUTHORIZATION or FOGGYKITCHEN_MCP_TOKEN.

• mcp/foggykitchen-stdio-bridge.mjs remains available as a local stdio bridge fallback and forwards tool requests to FoggyKitchen.

• skills/foggykitchen/SKILL.md teaches Codex or Claude Code when and how to use the plugin.

• scripts/list-tools.mjs is a direct HTTP smoke test using FOGGYKITCHEN_MCP_AUTHORIZATION or FOGGYKITCHEN_MCP_TOKEN.

Troubleshooting

• If Claude Code or Codex reports that the MCP server is not logged in, start the OAuth flow from the client: codex mcp login foggykitchen for Codex, or /mcp authentication for Claude Code.

• If the browser opens FoggyKitchen Settings instead of the OAuth approval screen, restart the client and start login again from the MCP client. Settings are for Personal MCP Tokens, not OAuth consent.

• If MCP tools are unavailable after installation, restart the AI coding client or start a new session so the server configuration is reloaded.

• If a Personal MCP Token was pasted into chat, shell history, logs, or source control, rotate it in FoggyKitchen Settings -> AI Developer Tools.

About

FoggyKitchen MCP connects Codex, Claude Code, and compatible AI coding agents to curated Azure, OCI, Terraform/OpenTofu, Kubernetes, and cloud architecture context from FoggyKitchen.

Learn more about FoggyKitchen MCP: https://foggykitchen.com/mcpserver