Don't want to enter an API key? Bring the AI you already pay for. MCP (Model Context Protocol) is now the industry-wide open standard — governed by the Linux Foundation and adopted by Anthropic, OpenAI, Google, Microsoft, xAI, and more — so Claude, ChatGPT, Copilot, Grok, Gemini, and any other MCP-capable client can drive Just Signal directly. Your subscription's model does the research and writing; Just Signal supplies what it uniquely owns — live Day-0 price grounding, mechanical dollar-target math, transmission stamps, publishing to the public library, and the Signal Scorecard. No API key is ever entered, and generation costs you nothing beyond the subscription you already have.
Web-based AI apps can't run local programs, so they connect to our hosted MCP endpoint. Add this URL wherever your AI app configures connectors:
| Your AI | Where to add it | Status |
|---|---|---|
| Claude (claude.ai / Desktop) | Settings → Connectors → Add custom connector → paste URL | ✅ Remote connectors (Pro/Max/Team/Enterprise) |
| ChatGPT | Settings → Connectors (enable Developer Mode) → New connector → paste URL | ✅ Plus / Pro / Business / Enterprise |
| Microsoft Copilot | Copilot Studio → your agent → Tools → Add MCP server → paste URL | ✅ Copilot Studio (consumer Copilot rolling out) |
| xAI Grok | grok.com → Connectors → add remote MCP → paste URL | ✅ Paid Grok accounts |
| Google Gemini | Gemini CLI: add an httpUrl MCP entry · Gemini Enterprise: admin connectors | ✅ CLI & Enterprise · consumer app in beta |
| Meta Llama | No MCP in the Meta AI app — run Llama in an MCP-capable client (LM Studio, Ollama + an MCP client) and use Option A or B | ⚠ Via open clients |
| Perplexity / Mistral / others | Their connector settings → paste URL | ✅ Where MCP connectors are offered |
The hosted endpoint exposes all four tools openly, including publish_report. Community publishing needs no key and no token: the server itself grounds every Day-0 price and renders every page, so the only thing anyone can add to the library is a genuine Just Signal report. To keep it fair, tokenless publishing is rate-limited (a few reports per connection per day, plus a global daily budget); read tools are never limited.
Clients that can launch local programs get the full experience, including publishing, from a tiny package that runs on your machine:
⬇ DOWNLOAD JUST-SIGNAL-MCP.ZIP| Tool | What it does |
|---|---|
| get_stock_quote | Live price + % change via Just Signal's market-data relay — establishes a ticker's Day-0 baseline. |
| publish_report | The big one: grounds every ticker's Day-0 price, computes exact dollar targets from your Claude's percentage projections, renders the full report (with Go-Ahead / Over-&-Out / Time-Out stamps and Scorecard data), and publishes it to Past Transmissions. Returns the live URL. |
| list_reports | Browse the public library. |
| get_scorecard | Audit past signals — projected vs. actual, graded against live prices. |
cd just-signal-mcp
npm installclaude_desktop_config.json{
"mcpServers": {
"just-signal": {
"command": "node",
"args": ["/ABSOLUTE/PATH/TO/just-signal-mcp/index.js"]
}
}
}claude mcp add just-signal -- node /ABSOLUTE/PATH/TO/just-signal-mcp/index.jsget_stock_quote to check each Day-0 price, then publish_report."get_scorecard any time to see how past calls are holding up.Targets must be percentages ("+18%") or multiples ("2.5x") — never dollar figures; the server computes dollars from the live Day-0 price so the arithmetic is real. Private companies or no ticker? Use (PRIVATE) and that trend honestly displays relative basis. Published reports are public — they join the same library, sitemap, and Scorecard as every other transmission.