Connectors - Chance

Chance runs a hosted (remote) MCP server so AI assistants can call verify_intent / get_verification over the network — no local process to install.

https://harness.chance.cc/api/mcp

Two ways to authenticate, depending on the client:

One-click OAuth — Claude.ai, Claude Desktop, Claude mobile, and ChatGPT open a Chance sign-in screen; no key to paste.
API key — the Anthropic and OpenAI developer APIs pass your chance_sk_ key as a header.

Prefer a local install (Claude Desktop / Cursor / Windsurf via npx)? Use the stdio MCP server instead — same tools, runs on your machine.

Claude.ai · Claude Desktop · Claude mobile (OAuth)

Open connector settings

Settings → Connectors → Add custom connector (Team/Enterprise: an owner adds it under Organization settings).

Add the URL

Paste https://harness.chance.cc/api/mcp and click Add. No OAuth client ID/secret needed — Chance registers your client automatically.

Click Connect and approve on the Chance screen (sign in with the same account as your dashboard). Done — verify_intent is now available in chat.

ChatGPT (OAuth)

Enable Developer Mode (Settings → Connectors → Advanced), then Add custom connector and enter https://harness.chance.cc/api/mcp. Approve the Chance sign-in. Developer Mode exposes the full read/write tools (verify_intent is a write tool); the read-only Deep Research connector path is not the one you want here.

Anthropic Messages API (key)

Call the remote server directly from the API with the MCP connector — your chance_sk_ key goes in authorization_token:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "anthropic-beta: mcp-client-2025-11-20" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-4-8",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Before trading, verify my order against my mandate."}],
    "mcp_servers": [{
      "type": "url",
      "url": "https://harness.chance.cc/api/mcp",
      "name": "chance",
      "authorization_token": "chance_sk_live_..."
    }]
  }'

The MCP-connector beta header and exact fields evolve — see Anthropic’s MCP connector docs for the current version.

OpenAI Responses API (key)

Add a remote MCP tool and pass your key as a header:

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "model": "gpt-5",
    "input": "Before trading, verify my order against my mandate.",
    "tools": [{
      "type": "mcp",
      "server_label": "chance",
      "server_url": "https://harness.chance.cc/api/mcp",
      "headers": { "Authorization": "Bearer chance_sk_live_..." },
      "require_approval": "always"
    }]
  }'

Any MCP client, via bridge (key)

For a client that only speaks local stdio, bridge to the hosted server with your key:

npx mcp-remote https://harness.chance.cc/api/mcp \
  --header "Authorization: Bearer chance_sk_live_..."

How the OAuth flow works

Chance implements the MCP Authorization spec (OAuth 2.1 + PKCE):

Discovery — /.well-known/oauth-protected-resource and /.well-known/oauth-authorization-server
Dynamic Client Registration — clients register automatically; nothing to configure
PKCE (S256), single-use authorization codes, and rotating refresh tokens
The consent screen authenticates you with the same identity as the dashboard, then issues the client a scoped token that spends your credits

Nothing to set up on your side — just add the URL and approve.

​Claude.ai · Claude Desktop · Claude mobile (OAuth)

​ChatGPT (OAuth)

​Anthropic Messages API (key)

​OpenAI Responses API (key)

​Any MCP client, via bridge (key)

​How the OAuth flow works

Claude.ai · Claude Desktop · Claude mobile (OAuth)

ChatGPT (OAuth)

Anthropic Messages API (key)

OpenAI Responses API (key)

Any MCP client, via bridge (key)

How the OAuth flow works