Documentation

AI and MCP

What Mango AI does inside the product, and how to connect your own assistant to the same operations.

Mango AI proposes, it does not act

Mango AI reads an incident and proposes a cause and a fix; a person approves, it never acts.

The MCP endpoint

MCP server URL
https://api.hopmango.com/mcp

Your assistant signs in with OAuth 2.1 and asks your permission in the portal; there is no key to paste.

What the token may do is capped by the role of whoever approved it; destructive calls and paging a person by hand need a confirmation at the time.

Connecting a client

Claude Code, in a terminal
claude mcp add --transport http hopmango https://api.hopmango.com/mcp

The first call opens the browser for the authorization; after that the client keeps its own token.

Claude Desktop, Cursor and other MCP clients
{
  "mcpServers": {
    "hopmango": {
      "type": "http",
      "url": "https://api.hopmango.com/mcp"
    }
  }
}

Add this to the client configuration file; no credential goes in it.

Skills

The tools say what can be done; this one says how to do it well.

One Skill, hopmango, in six chapters: onboarding a project, alert hygiene, synthetic monitoring and CDN triage, real user monitoring, diagnosing an incident, and on-call. Your assistant loads it when the conversation calls for it, and it is a single file at https://hopmango.com/dl/skills/hopmango/SKILL.md.

For Claude Code
mkdir -p ~/.claude/skills
curl -fsSL https://hopmango.com/dl/skills/skills.tar.gz | tar -xz -C ~/.claude/skills

That is where Claude Code reads personal Skills from. Nothing needs restarting.

For any other assistant
curl -fsSL https://hopmango.com/dl/skills/skills.tar.gz | tar -xz -C <your skills directory>

Extract it wherever your client reads Skills from. Checksums are listed in https://hopmango.com/dl/skills/index.json.

The file on its own
curl -fsSL --create-dirs -o ~/.claude/skills/hopmango/SKILL.md \
  https://hopmango.com/dl/skills/hopmango/SKILL.md

Prompts

A client that cannot install Skills asks for the same material by name, over the protocol: instrument_project, diagnose_incident, review_alert_hygiene, draft_postmortem.

Everything an assistant does over MCP is available over REST too: API documentation