Guide
MCP vs API: which to choose.
You bring your own AI, two ways. Drive Familiar from a Claude, ChatGPT, Gemini, or Grok subscription you already pay for, or paste a provider API key into the built-in chat and pay per token. Both routes reach the same tools, so the choice comes down to cost and how you play.
Which should you pick?
The short version: pick MCP if you already pay for Claude, ChatGPT, Gemini, or Grok and play most weeks, because the plan you hold already covers the AI and nothing extra is billed per token. Pick an API key if you want a specific model, tighter control over spend, or you hold no subscription to lean on. For nothing per token at all, point the built-in chat at a model running on your own hardware.
The subscription route is not unlimited. It spends your plan's normal message and usage quota, so a marathon night can still hit its rate limits. It adds no per-token bill on top, but the usage is never free.
| Factor | MCP (your subscription) | API key (built-in chat) |
|---|---|---|
| Who bills you | Your chat plan, no per-token add-on | Your provider, per token |
| Cheaper when | You play most weeks | Light or occasional play |
| AI cost | The flat plan you already pay | Roughly $0.50 to $3 a session |
| Setup | Add a connector, the wizard writes it | Paste a key and start typing |
| Model choice | What your plan includes | Any model the provider offers |
| Offline | No, the subscription runs online | Yes, pointed at a model you run |
| Players asking Familiar in the table chat | Your client answers when you say "check the table" | Answered automatically |
| Chat-only extras | Not over MCP | Transcription, dictation, NPC auto-pilot (D&D 5e), slash commands, drag-and-drop, the Art Library, the Edit button on an image |
Both paths reach the same tools. What differs is who you pay, and a short list of conveniences the built-in chat runs on its own.
What does each one actually cost?
Start with the part that does not change. Familiar itself is $6 a month, or $48 a year, with the first 14 days free, the same whichever route you pick. On top of that sits the AI cost, and that is where the two split.
On a subscription, the AI is already paid for. A Claude, ChatGPT, Gemini, or SuperGrok plan is a flat monthly price, and connecting it over MCP adds no per-token charge on top. Over a run of weekly sessions, that flat price usually buys more play than metered tokens do.
On a provider key, you pay per token, in and out, for every message. A single session runs roughly $0.50 to $3, depending on the model. Rates move, so check current provider pricing before you commit. Light play stays cheap; volume is where it adds up.
Or pay nothing per token at all: point the built-in chat at a model on your own hardware, and no provider bills you. Either way the AI cost is yours directly. Familiar never marks up or proxies your tokens.
One thing trips people up: a subscription and the API are billed apart. A ChatGPT plan does not hand you API credits, and an API balance does not unlock the chat app. You are choosing one route or the other, not topping up a shared pool.
When is MCP the cheaper route?
MCP is the easy call when you already pay for the subscription. Claude, ChatGPT, Gemini, or Grok is on the bill either way, so pointing it at your Foundry game adds nothing. You need a plan that exposes MCP connectors; the connect guide lists which ones per assistant.
Three of the four connect through the maker's own agent: ChatGPT through Codex, which is OpenAI's, Gemini through Antigravity, which is Google's, and Grok through Grok Build, which is xAI's. Either way you are talking to your usual assistant, only over MCP.
Familiar's tool catalogue spends less of that plan than its size suggests: no client reads every tool description at connect. The capabilities page covers what it does cost.
- You already pay for Claude, ChatGPT, Gemini, or Grok. The AI cost is covered, and the connector puts it to work at the table.
- You play regularly. A flat monthly price suits a weekly game better than a meter that climbs with every session.
- You want a predictable cost. One subscription, one known number each month, with no per-token surprises after a long night.
- You are happy inside your plan's limits. A typical session fits; a long one can brush the plan's rate cap.
When is an API key the better call?
The API key wins when you want control, or when you hold no chat subscription to lean on. It is the simplest way to start: pick a provider, paste a key, and talk to your game from Foundry's sidebar. The conveniences that live in that chat window only are in the comparison table above.
- You want a specific model your subscription does not include. A key reaches the full provider catalogue, not just what a chat plan ships.
- You want fine control over the model and the spend, paying only for the tokens a session uses.
- You have no chat subscription. With light or occasional play, a per-token bill can come in under a monthly plan you would barely use.
- You run a headless or automated setup where a desktop chat app is not in the picture.
- You want to play offline. A model on your own machine answers through this same chat, with no provider in the path and no token bill.
Which model should you run?
One more lever sits on top of the route: the model. A bigger model reasons better over a messy fight but costs more per token, and how hard it thinks costs tokens too. Whichever route you pick, you choose the model and can change it whenever you like. Picking one is its own guide.
Pick the cheaper route and start
If you already pay for Claude, ChatGPT, Gemini, or Grok and play most weeks, connect over MCP and let the plan you hold do the work. If you want a particular model, tighter control, or you have no subscription to lean on, paste an API key and pay for what you use. Neither is locked in; you can switch routes later without redoing your prep.
Install Familiar in Foundry to get going. If you are weighing the two against your own habits, the Discord is a good place to ask.
More in Connect your assistant
Claude in Foundry: Desktop or Code
NewAnthropic builds both clients, so no agent app sits in the path. Desktop and Code differ in how you add each one, how many tools each carries, and whether a fetched picture ever reaches your screen.
ChatGPT in Foundry via Codex
NewThe Chat tab is not the piece that connects. The Codex tab in the ChatGPT app is, on every plan including Free, and so is Codex CLI from Plus upward: OpenAI's own agent either way. Both read the same config file.
Gemini in Foundry via Antigravity
NewAntigravity is Google's agent, and it is what reaches your Foundry game. Any Google account gets you in, including a free one. Or skip the agent altogether: a Google API key, pasted into Familiar's own chat window.
Grok in Foundry via Grok Build
NewxAI's terminal agent is the one Grok that speaks to a Foundry game, on the SuperGrok plan you already pay for. One command registers Familiar, and grok mcp doctor familiar proves the link before you type a word. The Grok app itself cannot reach a server on your machine.
Choosing an AI model
NewWhich model, and how hard should it think? A capable model with reasoning headroom is the second lever behind good prep. Turn thinking up for fiddly rules, down to keep replies fast.
Running a local LLM
NewRun the model on your own PC instead of a provider, so your world never leaves the machine and the bill moves from a token meter to your graphics card. Setup is one settings tab and one switch. Without a discrete GPU, expect minutes per turn.
New to Familiar? I'm Ryan, the person who built it. The Discord is small and brand new, so if you join now I'll help you get set up myself.