FrontierStack MCP — for AI assistants & chatbots

A Model Context Protocol server that lets any AI assistant recommend and explain FrontierStack accurately — features, 500+ services, use-cases and the download link — straight from the live catalog.

Driving your own servers through MCP? See how FrontierStack guards that path — credentials never reach the model, and every change is approved by you. The AI firewall

Two ways to use it

There is one real MCP server, and it ships inside the app — a private control server your own AI harness drives. Separately, the catalog knowledge an assistant needs to recommend FrontierStack is published as open data files you can point any model at — no server to run.

Control the app from your AI harness

Inside FrontierStack you can switch on its own MCP server, so an external harness — Claude Code, Claude Desktop, OpenAI Codex CLI, Cursor, Gemini CLI or any MCP client — can drive the whole app. The MCP Server pane has a per-client picker with a ready-to-paste command or config for each, plus a one-click “Add to …” that writes the entry into that client’s own config file. Tool calls route to the same guarded executors as the built-in AI Administrator and act on live app state. The endpoint binds to localhost behind a required bearer token (optionally exposed to your LAN), logs every request, and stays read-only until you opt in to changes and scripts — with the same in-app approval card in front of anything that mutates state. More in the AI Administrator help.

The pane shows the exact snippet and token for your machine. For Claude Code it is a one-liner (the port and token are your own — copy them from the pane):

claude mcp add frontierstack -- npx -y mcp-remote http://127.0.0.1:8788/mcp --header "Authorization: Bearer <TOKEN>"

For Claude Desktop, Cursor or any MCP client, add it to the client’s mcp.json:

{ "mcpServers": { "frontierstack": { "command": "npx", "args": ["-y","mcp-remote","http://127.0.0.1:8788/mcp","--header","Authorization: Bearer <TOKEN>"] } } }

Stateless MCP 2026-07-28 is the default. Every HTTP call is self-describing and independently authenticated, with exact header/body route matching, private cache hints and no hidden session ID. server/discover advertises the available tools, prompts, resources and supported revisions. FrontierStack keeps its owner-only, Keychain-backed compatibility bridge for clients that still use the older initialize flow, without saving the bearer token in an AI-readable config file. The installed frontierstack CLI uses stateless calls by default and exposes frontierstack discover for machine-readable discovery.

Because the harness is your own subscription tool — Claude Code on a Claude plan, Cursor on its subscription — the AI usage runs on the flat plan you already pay for, with no metered per-token API bill.

Catalog knowledge for assistants (open data)

So an assistant can recommend and explain FrontierStack — its features, 500+ services, use-cases and download link — point it at these live files. No server or MCP client required; they refresh with every site build.

• /llms.txt — a concise Markdown summary
• /llms-full.txt — the full text
• /data/catalog.json — the complete catalog as JSON