Free download · Open source · 5MB

FoxAPIs Desktop. Connect once, use everywhere.

A tiny desktop app that runs a local Model Context Protocol relay on your own machine. Save your API key once, point every AI client at one local URL, and watch every call stream into a single live activity log.

http://127.0.0.1:8732/mcp
Download for macOS universal Download for Windows x64 Download for Linux AppImage

Or via Homebrew: brew tap OsakaSaul/foxapis && brew install --cask foxapis-desktop

Why a desktop relay?

Web-based MCP works, but every client needs its own auth dance. Pasting bearer tokens into Claude Desktop, then again into Cursor, then again into n8n. Rotating the key? Now do that three times. Wonder which client just burned credits on a bad call? Check three different log surfaces. FoxAPIs Desktop fixes all of that with one local hop.

One key, every client

Save your FoxAPIs key once in the desktop app. Every connected client points at the same local URL. Rotate the key in one place and every client follows.

Wired in v0.1

Live activity log

Every MCP call across every client streams into one timeline. Tool name, originating client, status, latency, cache hit or miss, errors. Filter by tool, filter by client, find the bad call.

Wired in v0.1

Tool catalog and sandbox

Every FoxAPIs MCP tool, browsable, with full descriptions, input schemas, and a "Try it" panel. Run a tool, see the response, copy the args into your real client config.

Wired in v0.1

Connector hub

One-click setup for Claude Desktop, Cursor, and Continue. The app detects each config file, merges in the FoxAPIs entry, and asks before saving. Other clients render copy-paste snippets.

Wired in v0.1

Offline read

Cached dossiers and reports work without the network. Per-request hash cache lives in your app cache directory. Useful on planes, useful for repeat reads, useful for keeping live API costs down.

Read-only in v0.1

System tray balance

Current API balance plus 24-hour and 7-day burn rate, glanceable from the menubar. No need to flip to the dashboard mid-flow when you want to know what a workflow is costing you.

Scaffold — see roadmap

One URL, then any client

The desktop app spawns a tiny HTTP server on 127.0.0.1:8732 as soon as it launches. Point any MCP-aware client at it. The relay forwards every request to FoxAPIs in the cloud with your bearer key attached. Latency is whatever your network is, plus a microsecond of localhost hop. Drop the same three lines below into any client config and you are done.

# Claude Desktop — claude_desktop_config.json
{
  "mcpServers": {
    "foxapis": {
      "url": "http://127.0.0.1:8732/mcp"
    }
  }
}

# Cursor — ~/.cursor/mcp.json
{
  "mcpServers": {
    "foxapis": { "url": "http://127.0.0.1:8732/mcp" }
  }
}

# Continue — ~/.continue/config.json
{
  "experimental": {
    "modelContextProtocolServers": [
      { "name": "foxapis",
        "transport": { "type": "http",
                       "url": "http://127.0.0.1:8732/mcp" } }
    ]
  }
}

Or just hit Install in the Connector Hub tab inside the app. The desktop app does the file edit for you, with a confirmation dialog before saving. If a config already exists, the merge is non-destructive — your other servers stay where they are.

Works with eight clients out of the box

Anything that speaks the Model Context Protocol works with the local relay. The eight below have first-class support in v0.1 — three with one-click writers, the rest with copy-paste snippets generated for your platform. New clients keep landing as the protocol spreads, and the catalog inside the app updates automatically.

Claude Desktop Cursor Continue Goose n8n LibreChat Open WebUI ChatGPT custom GPT

For the full integration matrix with status badges and walkthroughs, see the FoxAPIs Integrations Hub. Every client there can point at the local desktop URL or the hosted endpoint — same tools, same responses, your choice of hop.

Every tool, one connector

The FoxAPIs MCP server exposes a vetted set of research, scanning, visibility, and pipeline tools. The desktop app surfaces every one of them in the Tools tab with full descriptions, schemas, and a live "Try it" sandbox. Whatever the catalog grows to, the app pulls the live tool list on launch — anything new shows up in your local catalog within minutes of shipping.

Research

Person enrichment, dossier lookups, entrepreneur reports, investor reports, comparative analysis between two subjects, recent research history. The full research stack, callable as plain language from any client you connect.

Scanning and discovery

Mention scans across 40+ surfaces, journalist discovery filtered by beat or coverage angle. Run them headless from a workflow tool, or interactively from your editor with cached results visible in the desktop activity log.

Pipeline

Save subjects to your watch pipeline, subscribe to weekly digests, run deep-dive briefings on saved companies, and define custom pipeline watches that match your specific deal flow or coverage criteria.

GEO and AEO visibility

Score how well your domain shows up in answer-engine results from ChatGPT, Perplexity, and Gemini. Run audits, list top remediation actions, and track score deltas as you ship fixes — all from inside any AI client.

Reporting

Generate downloadable white-label reports in HTML, PDF, or plain text from cached data. Optional agency-branding override means you can ship a deliverable to a client without the FoxAPIs name on it if your business needs that.

Catalog auto-updates

The MCP server adds tools faster than the docs ship. The desktop app pulls the live tool list on launch, so anything new lands in your catalog and Try-it sandbox automatically. Browse the current catalog at /tools/.

Where your data goes

Every MCP call you make is forwarded directly to the FoxAPIs cloud endpoint with your bearer key attached, and the response is sent straight back to your local client. The desktop app does not phone home, does not log to any third party, and does not touch any other network. Your bearer key is stored in your operating system's keychain — macOS Keychain, Windows Credential Vault, gnome-keyring or libsecret on Linux. The app never writes your key to disk in plain text.

Activity-log events live in memory and are cleared when you quit. Cached responses live in your app cache directory and stay there until you delete them, with a one-click "purge cache" in Settings. The whole codebase is open source: read it, audit it, fork it. github.com/OsakaSaul/foxapis-desktop.

Who this is for

If you are a solo developer wiring up Claude Desktop, Cursor, and a side n8n flow against the same data layer, the desktop app saves you three rounds of token-pasting and gives you one log to grep when something goes sideways. If you are an agency running multiple clients through workflow tools, the desktop relay is the cleanest way to keep keys out of vendor configs and rotate them on demand. If you are an indie hacker building agentic workflows, this is the missing piece between "I have an API key" and "I have ten clients calling the same key with one log to look at." Most people who install the app are using at least three MCP-aware clients before the week is out, and the activity log becomes the surface they keep open while building.

The relay is free. You bring your own FoxAPIs key — see pricing for the tiers, or grab a trial key with no card on file from /trial/. The cloud endpoint is also documented at /docs/ if you would rather skip the desktop app entirely and call the hosted endpoint directly. Both paths hit the same tools and return the same shape of response, so the choice is really about whether you want a single local log surface and a key vault, or whether you would rather keep things plain and call the hosted URL from each client config separately.

There is no lock-in. Uninstall the app and your existing client configs keep working — they just point back at the hosted endpoint instead of the local hop. The configs the app writes are plain text files, edited in place with a backup written next to them on first save. Nothing the desktop app touches is proprietary. Read the source, copy the relay logic into your own tooling, fork the repo and ship a different UI on top of the same protocol. The only thing the desktop binary does that you cannot trivially do yourself is the auth-merge dance across config formats, plus the activity log overlay across clients.

Common questions

Does it slow my calls down?

The localhost hop adds well under one millisecond on every machine we have measured. Total response time is dominated by the cloud round-trip exactly the same way as a direct call from your client to the hosted endpoint. Cache hits return locally in microseconds.

What happens if I am offline?

Cached reads still work. Anything not in cache returns a clean error to the client. The desktop app surfaces the offline status in the system tray and pauses the activity-log writer to in-memory only, so you do not pollute the log with retry storms.

Can I run it headless on a server?

v0.1 is a desktop binary with a UI. v0.3 will ship a headless server mode for team and agency setups. Until then, point your hosted clients directly at the cloud endpoint at mentionfox.com — same tools, same shape, no relay needed.

Why a separate desktop app and not a browser tab?

Browser sandboxes cannot edit your editor configs, cannot bind a local port reliably across operating systems, and cannot read your OS keychain. Native is the only honest path for a one-click connector hub. Plus the binary is 5MB.

Is this only for MentionFox subscribers?

The desktop app is fully usable with any FoxAPIs key. MentionFox subscribers get extra research tools surfaced automatically. Free-tier developer keys also work — see pricing for what each tier unlocks at the API layer.

How do updates work?

The app checks for a new release on launch and shows a non-blocking banner if one is available. Updates are signed and ship through the same GitHub Releases page you downloaded from. You can also subscribe to the release feed for changelogs without auto-update prompts.

Roadmap

v0.1 ships the architecture. v0.2 and beyond fill it in. Issues and pull requests welcome on the GitHub repo — this is a tool built in the open, with a public changelog and a public issue tracker.

v0.2 — next up

Real credit balance in the system tray (per-response credit metadata from the MCP server). Full one-click writers for Goose, n8n, LibreChat, and Open WebUI. Multi-account switcher for agencies running more than one workspace. Offline write replay queue for actions that need to land when the network comes back. Native menubar widget on macOS.

v0.3 — soon after

Local model fallback for cache-only reads — re-ask a cached dossier with a different framing without touching the cloud. Workflow recorder so you can replay a common sequence of tool calls with a single shortcut. Team and agency mode where one desktop instance serves multiple authenticated users via an OAuth pivot.