Install

Add InferSports to your agent.

Three ways to connect, all keyless on the Free tier: a remote MCP server (one line), a bundled agent Skill, or plain REST. Pick your host below for the exact setup.

One line · any agent
paste into your agent
Read https://infersports.dev/skill.md and follow the instructions to connect to InferSports.

The agent fetches the guide and picks the right surface for its host.

Fastest · MCP host

Claude Code, one command:

bash
claude mcp add --transport http \
  infersports https://api.infersports.dev/mcp
Fastest · Skill host

Let the agent install itself — paste one prompt and it clones, copies the Skill into the right folder, and verifies it for you.

Get the install prompt ↓
Pick your path

Three connect surfaces.

① Remote MCP · recommended

Connect over MCP.

One keyless endpoint works in every MCP host. The field name and root key differ by host — watch for httpUrl (Gemini CLI), serverUrl (Windsurf) and the servers root key (VS Code).

remote MCP endpoint
# keyless — works in every MCP host
https://api.infersports.dev/mcp
Claude Code
claude mcp add --transport http \
  infersports https://api.infersports.dev/mcp

one-line CLI

Cursor
{
  "mcpServers": {
    "infersports": { "url": "https://api.infersports.dev/mcp" }
  }
}

~/.cursor/mcp.json (or .cursor/mcp.json)

OpenAI Codex
[mcp_servers.infersports]
url = "https://api.infersports.dev/mcp"

# or one line:
# codex mcp add infersports --url https://api.infersports.dev/mcp

~/.codex/config.toml

OpenCode
{
  "mcp": {
    "infersports": { "type": "remote", "url": "https://api.infersports.dev/mcp", "enabled": true }
  }
}

opencode.json

Gemini CLI
{
  "mcpServers": {
    "infersports": { "httpUrl": "https://api.infersports.dev/mcp" }
  }
}

~/.gemini/settings.json — note: httpUrl

Windsurf
{
  "mcpServers": {
    "infersports": { "serverUrl": "https://api.infersports.dev/mcp" }
  }
}

~/.codeium/windsurf/mcp_config.json — note: serverUrl

VS Code (Copilot)
{
  "servers": {
    "infersports": { "type": "http", "url": "https://api.infersports.dev/mcp" }
  }
}

.vscode/mcp.json — note: root key is "servers"

Claude Desktop / Cowork
Settings → Connectors → Add custom connector
URL:  https://api.infersports.dev/mcp
(keyless — no API key needed)

UI — custom connector

OpenAI Agents SDK
HostedMCPTool(tool_config={
  "type": "mcp",
  "server_label": "infersports",
  "server_url": "https://api.infersports.dev/mcp",
  "require_approval": "never",
})

Python / TypeScript

Any MCP client
{
  "mcpServers": {
    "infersports": { "url": "https://api.infersports.dev/mcp" }
  }
}

generic config

② Agent Skill

Bundled verbs for small-context and local agents.

A folder with a SKILL.md and twelve read-only verbs (today, events, digest, match, preview, line, fair, scoreprob, compare, scan, convert, result). Keyless, deterministic, and tiny on context — source on GitHub. The fastest install is to let the agent do it:

agent · install prompt
Install the "InferSports" sports-odds skill for me, verify it works, then show me what to ask.

0. If <skills_dir>/infersports already exists this is an UPDATE: compare
   cat <skills_dir>/infersports/VERSION with the latest
   (curl -s https://infersports.dev/skill.tgz | tar -xzO infersports/VERSION);
   if they differ — or I asked for a reinstall — rm -rf <skills_dir>/infersports and continue.
1. Find where your agent loads Agent Skills (create the folder if missing):
   • Claude Code → ~/.claude/skills/   (or ./.claude/skills/ for this project only)
   • OpenAI Codex → ~/.codex/skills/
   • OpenClaw → ~/.openclaw/workspace/skills/
   • other → your agent's skills directory
2. Download the skill bundle (the no-git install — it contains SKILL.md + scripts):
   curl -sSL https://infersports.dev/skill.tgz -o /tmp/infersports-skill.tgz
3. Extract it into your skills directory; it unpacks to a folder named infersports:
   tar -xzf /tmp/infersports-skill.tgz -C <skills_dir>
4. chmod +x <skills_dir>/infersports/scripts/*.sh
5. Verify offline:  INFERSPORTS_MOCK=1 <skills_dir>/infersports/scripts/today.sh --limit 3
6. Verify live (keyless, no signup):  <skills_dir>/infersports/scripts/today.sh --limit 3
7. Reload skills if needed, confirm the install, and give me 3 example questions to try.

Where each host loads skills

HostSkills directory
Claude Code~/.claude/skills/— or ./.claude/skills/ for one project
OpenAI Codex~/.codex/skills/
OpenClaw~/.openclaw/workspace/skills/— one-line ClawHub install coming ~Jun 22
Hermes~/.hermes/skills/— or external_dirs in config.yaml

Prefer to do it by hand? Download infersports.dev/skill.tgz and extract it into the directory above (tar -xzf skill.tgz -C <skills_dir>) — it unpacks to an infersports/ folder — then chmod +x infersports/scripts/*.sh.

Updating or reinstalling is the same install run again — the tarball always carries the latest release. Check what you have with cat <skills_dir>/infersports/VERSION; the latest published version is tar -xzO infersports/VERSION from a fresh skill.tgz.

③ Plain REST

No agent? Just HTTP.

Every MCP tool is also a plain REST endpoint. Keyless on the Free tier.

bash
# Keyless (Free tier) — no signup
curl https://api.infersports.dev/v1/events

# With an API key (higher limits + Pinnacle)
curl -H "Authorization: Bearer isk_…" \
  "https://api.infersports.dev/v1/events/<id>/odds?format=hk"

# Any MCP tool over plain REST — no MCP client needed
curl -X POST https://api.infersports.dev/v1/mcp/match_info \
  -H "Content-Type: application/json" \
  -d '{"query": "France vs Argentina"}'
Try it

Ask it anything.

However you connected, just ask in plain language. A real, keyless call comes back with live numbers — no account, no card. A few to start with:

ask your agent
› Who's favored in tonight's matches?
  France vs Argentina — France 58% (AH −0.25), kickoff 20:00 local.

› What's the sharp Asian handicap on Man City vs Arsenal, in HK odds?
› How has the line moved on Real Madrid vs Barcelona since it opened?
› Scan today's football slate for value.
› What was the score of Brazil's last match?
FAQ

Install questions.

What is the one-line install for any agent?

Paste: Read https://infersports.dev/skill.md and follow the instructions to connect to InferSports. The agent fetches the guide and picks the right surface (MCP, Skill, or REST) for its host. Keyless.

How do I add InferSports to Claude Code?

Run: claude mcp add --transport http infersports https://api.infersports.dev/mcp. It is keyless on the Free tier, so no API key is needed to start.

How do I add InferSports to Cursor?

Add it to ~/.cursor/mcp.json: { "mcpServers": { "infersports": { "url": "https://api.infersports.dev/mcp" } } }. Or use an Add to Cursor button.

How do I add InferSports to OpenAI Codex?

Add [mcp_servers.infersports] with url = "https://api.infersports.dev/mcp" to ~/.codex/config.toml, or run codex mcp add infersports --url https://api.infersports.dev/mcp.

How do I install the InferSports agent Skill?

Download https://infersports.dev/skill.tgz and extract it into your agent’s skills directory — for example: tar -xzf skill.tgz -C ~/.claude/skills (or ~/.codex/skills). It unpacks to an infersports/ folder with SKILL.md; then ask a question to confirm. In Claude Code you can instead run: /plugin marketplace add infersports/infersports-skill. Keyless.

Do I need an API key to use InferSports?

No. The remote MCP server and the Skill are keyless on the Free tier. An optional isk_ API key raises rate limits and unlocks the sharp book (Pinnacle).

Connected. Now ask it anything.

Live Asian odds, the sharp line, opening-line movement, and finished scores — read-only, keyless, free during the World Cup beta.