discord-mcp

Repository: github.com/sandraschi/discord-mcp

FastMCP 3.1 Discord MCP server with sampling (server-side Ollama / OpenAI-compatible or client LLM), agentic workflow (discord_agentic_workflow), skills (SkillsDirectoryProvider + skill://…/SKILL.md), prompts, and a 2026 fleet-style dashboard (React + Vite + Tailwind, dark glass shell).

• Backend: 10756 — REST (/api/v1/…) + MCP streamable HTTP at `/mcp`
• Frontend: 10757

1. Create a bot at Discord Developer Portal. Copy the bot token.
2. Invite the bot to your server (OAuth2 URL Generator, scope: bot).
3. Copy .env.example to `.env` in the repo root (discord-mcp/.env) and set DISCORD_TOKEN=... (no quotes). The backend loads this file on startup (python-dotenv). Restart the server after changing .env. If the token is already set in the system environment or Cursor MCP env, that value wins (.env does not override existing vars).
4. For local agentic sampling (when the MCP host does not provide sampling): run Ollama and optionally set DISCORD_SAMPLING_BASE_URL / DISCORD_SAMPLING_MODEL. Set DISCORD_SAMPLING_USE_CLIENT_LLM=1 if you want the host LLM to sample and the server handler only as fallback.

• Webapp (backend + dashboard): from repo root run .\start.ps1 or start.bat, or from webapp/: .\start.ps1.
• Backend only: uv run python -m discord_mcp.server --mode dual --port 10756
• Stdio (MCP only): uv run python -m discord_mcp.server --mode stdio

• This repo as workspace root: .cursor/mcp.json registers discord-mcp (stdio). Set `DISCORD_TOKEN` in that file’s env or rely on your shell environment. Reload MCP / restart Cursor after edits.
• Global / multi-root merge: copy the `discord-mcp` block from **`cursor-con