🐾 ForagePlan a trip

AI

What's running, what it costs, what it remembers — and how to use your own assistant instead.

The MCP zone

Put your travel agent inside your own AI

Forage and the Bearings commons both speak MCP — the standard way for an AI to use outside tools. So you don't have to come here at all: your own assistant can read what Forage knows about you, search real bear events and plan a trip with this site closed.

Why it matters

Event listings die with the apps they live in — an Instagram poster is unfindable in six months. Bearings keeps that data where an agent can reach it, and Forage adds the part that knows you.

The catch is where the thinking happens. The strongest models run in the cloud, so your trip and profile pass through someone else's computer. Local models are catching up fast, and the same connection works either way — you can move your privacy along with them without changing how you plan.

☁️ Cloud today
Strongest reasoning. Your data leaves your machine.
🏠 Local today
Nothing leaves your machine. Weaker at long plans, fine for lookups and drafts.
🏠 Local tomorrow
The bet: local closes the gap. Same connection, better privacy, nothing to migrate.

☁️ Cloud AI

Claude · ChatGPT — the easiest and most capable

The trade, plainlyEverything the AI reads goes to that company's servers — for Forage, that includes who you travel with, where you stay, your nights out and what you've told the desks. Their terms decide whether it's used for training, and those differ by plan and change over time. If you'd hesitate to tell a stranger, connect read-only or go local.

Forage in Claude

  1. In claude.ai (web or desktop), open Settings → Connectors and choose Add custom connector. On some plans it's under Settings → Integrations.
  2. Paste the Forage address and save:
    https://for4ge.com/mcp
  3. Claude sends you to Forage to sign in (with an email link if needed), then straight to the consent screen.
  4. Read the consent screen. It names Claude and what you're granting: forage:read (see your Forage) and forage:plan (start trips, set windows, talk to the desks — which uses your daily turns).
  5. Back in Claude, ask something only Forage knows: “What does Forage know about me?” It should read forage://me and come back with your crews, passport and travel windows.

Forage in ChatGPT

Same address, same consent. Custom connectors are under Settings → Connectors; on some plans turn on Developer mode first (Settings → Apps & Connectors → Advanced). Add https://for4ge.com/mcp as the server URL.

Bearings in the cloud

The commons is public and needs no account — no sign-in, no consent, nothing of yours attached. Add it the same way:

https://bearings.community/mcp

30-plus read-only tools: events, places, clubs, creators, title holders, inclusivity flags, press, date forecasts. Connect it everywhere — it's public community data, and it gets more useful the more agents can reach it.

🏠 Local AI

LM Studio · Ollama — nothing leaves your machine

The trade, the other wayYour profile never leaves your computer: no company's terms, no training, and it works on a plane. The cost is capability — local models are weaker at long multi-step plans, and small ones can fumble tool calls. It's the option that improves on its own with every good local release.

LM Studio (the easiest local route)

LM Studio speaks MCP natively — nothing extra to install.

  1. Sign in, then create an agent key at the top of this tab (it's shown once). Local apps let you set a header, so they use a key instead of the browser consent screen.
  2. In LM Studio, open the Program panel and choose Install → Edit mcp.json.
  3. Add both servers and save — Bearings needs no key:
    {
      "mcpServers": {
        "forage": {
          "url": "https://for4ge.com/mcp",
          "headers": {
            "Authorization": "Bearer frg_YOUR_KEY_HERE"
          }
        },
        "bearings": {
          "url": "https://bearings.community/mcp"
        }
      }
    }
  4. Load a model that's good at tool calling (a 7B+ instruct model with tool support) and approve the tools when asked.
  5. Ask it “read forage://me and tell me what you know about me” to confirm.

Ollama

Ollama alone can't do this — it runs models but can't call tools. Put a client on top, pointed at your local Ollama:

• Open WebUI — the usual choice. It reaches MCP through mcpo: run uvx mcpo --port 8000 -- <server> and point its Tools at http://localhost:8000.
• Goose or mcphost — command-line agents that speak MCP directly.
• Cherry Studio / LibreChat — desktop chat apps with MCP and Ollama support.

Whichever you pick, the two server addresses and the Authorization: Bearer frg_… header are the same as above.

Bearings locally

Even simpler — no key, no consent, no account: https://bearings.community/mcp. Worth adding even if you never connect Forage: it turns any local model into one that knows the bear calendar.

If you live in a terminal

Claude Code, Cursor and similar let you set headers directly — the quickest way to try it:

claude mcp add --transport http bearings https://bearings.community/mcp

claude mcp add --transport http forage https://for4ge.com/mcp \
        --header "Authorization: Bearer frg_YOUR_KEY_HERE"

How to tell it worked

What we'd recommend

As of September 2026. Forage speaks MCP over Streamable HTTP with OAuth 2.1 (or a bearer key for apps that set headers), up to protocol revision 2025-06-18. Bearings is public and read-only. Menu names in Claude, ChatGPT, LM Studio and Open WebUI move between releases — if a step doesn't match, look for “connectors”, “MCP” or “tools” in settings. The addresses above don't change.