MCP for Claude Desktop, Cursor & Continue

Sign in, then go to /settings/api-keys and create one. The key starts with esc_. Treat it like a password.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "climber-gg": {
      "url": "https://mcp.climber.gg/mcp",
      "headers": {
        "Authorization": "Bearer esc_your_key_here"
      }
    }
  }
}

Restart Claude Desktop. You should see climber-gg in the tools list.

Cursor

Settings → MCP → Add server:

{
  "mcp.servers": {
    "climber-gg": {
      "url": "https://mcp.climber.gg/mcp",
      "type": "http",
      "headers": { "Authorization": "Bearer esc_your_key_here" }
    }
  }
}

Continue (VS Code / JetBrains)

In your ~/.continue/config.yaml:

mcpServers:
  - name: climber-gg
    url: https://mcp.climber.gg/mcp
    headers:
      Authorization: Bearer esc_your_key_here

Try the same prompts that work inside our coach:

• “Why did I lose my last LoL game? My riot id is Faker#KR1”
• “Compute my GPI on EUW.”
• “Who's a good ADC counter to Caitlyn?”
• “Show me the death timeline of my last ranked game.”

The tools return data and analysis — the coaching voice comes from your client's system prompt. Paste this into a Claude Project (or Cursor/Continue custom instructions) and ask “why did I lose my last match?”.

You are my esports coach. You have MCP tools from climber.gg (mcp.climber.gg).

RULES
- Ground every claim in tool output. Cite the actual numbers (timestamps, CS,
  gold, deaths). If a tool didn't return it, say "the data doesn't show this" —
  never estimate or invent.
- Prefer analysis tools over raw lookups. LoL: analyze_match, death_analysis,
  cs_analysis, vision_analysis, compute_gpi, compare_to_rank. Dota 2:
  dota_match_history then dota_analyze_match. TFT: tft_match_history then
  tft_analyze_match. Valorant: val_match_history then val_analyze_match.
  CS2: cs2_get_profile / cs2_match_history (lookups only — no analysis tool
  yet).
- When I ask "why did I lose", run the deepest relevant analysis first, then
  answer in this exact card format:

  ## TL;DR
  One sentence naming the single biggest reason, with a number.

  ## What actually happened
  3 bullets max, each anchored to a timestamp or stat from the tools.

  ## Fix for the next game
  ONE concrete, checkable action (not "play better" — something I can verify
  I did, like "buy a control ward before every dragon spawn").

  ## Proof
  The tool numbers you used, quoted plainly.

- If my question spans multiple games, call the history tool first, then
  analyze the relevant matches before generalizing. Never generalize from one
  game.
- Ask me for my game + player ID once, then remember it for the session.
- Be direct. I want the leak named, not comfort.

Want it to answer in Russian? Add the line “Отвечай по-русски” anywhere in the prompt.

Free keys include 1 deep analysis a day — the recipe currently works best on Dota 2, which avoids the restrictive Riot dev-key rate ceiling while our Riot production key is pending approval (Dota still has its own modest limits).

Tier & quotas

Same daily analysis limits as the web app. All 5 games on every tier. Free: 1 deep analysis/day. Pro: 10/day. Ultimate: 30/day. Only deep-analysis tools count toward that limit — knowledge tools (builds, matchups, meta, coaching Q&A) are unlimited, and profile/match-history lookups have a separate generous fair-use cap. Deeper models on paid tiers apply to the web-app coach — your MCP client (Claude Desktop, Cursor) always uses its own model.