Install & connect

Two pieces connect your agent to the platform: an MCP server entry in your client's configuration, and the botz-mcp client skill that teaches your agent the consumption workflow.

1. Connect your MCP client

The platform speaks MCP over streamable HTTP. IDE and desktop clients need only the server URL — https://mcp.botz.ai/mcp (or your own deployment's /mcp). The first time the client connects it opens a browser window to sign in with GitHub (MCP OAuth). There is nothing else to configure: no tokens, headers, or environment variables.

Cursor

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global), or install the Botz Skills plugin, which adds this entry for you:

{
  "mcpServers": {
    "botz": { "url": "https://mcp.botz.ai/mcp" }
  }
}

Claude Code

claude mcp add --transport http botz https://mcp.botz.ai/mcp

Then run /mcp and choose Authenticate to sign in with GitHub.

VS Code & other MCP clients

Add an HTTP MCP server named botz with the URL https://mcp.botz.ai/mcp (in VS Code: MCP: Add Server… → HTTP, or .vscode/mcp.json with {"servers": {"botz": {"type": "http", "url": "https://mcp.botz.ai/mcp"}}}). Any client with MCP OAuth support signs in the same way. Clients that support elicitation get the best experience — the platform can ask clarifying questions mid-request when creating new skills.

2. Install the client skill

From the Cursor marketplace plugin

Add the zotoio/zoto-agents marketplace and install the zoto-botz-skills plugin (plugin docs). It carries the client skill and a botz MCP entry for https://mcp.botz.ai/mcp; your client signs in with OAuth the first time it connects, so there are no tokens to set. After that, just ask your agent for work. For a self-hosted server, override the entry's url in your own MCP config.

Manually (any client)

The skill is served by the platform itself. With the server connected, call skill_get botz-mcp (raw=true) from your client and save the body as:

.cursor/skills/botz-mcp/SKILL.md      # Cursor
.claude/skills/botz-mcp/SKILL.md      # Claude Code

Easier: once the MCP server is connected, tell your agent "fetch the botz-mcp skill with skill_get and install it into my skills directory" — it can bootstrap itself.

3. Verify

• Your client lists 12 skill_* / agent_* tools for the server.
• Ask your agent: "search the skill corpus for summarising documents" — you should get ranked matches back.

Headless & CI (API keys)

1. An admin opens the management console at /admin → IAM · API tokens and mints a key bound to a user label, role, and teams. The key starts with bzt_ and is shown once.
2. Store it in your CI secret store (for example a GitHub Actions secret named BOTZ_API_KEY). Never commit it or paste it into chat.
3. Send it as an Authorization: Bearer header on every request to /mcp.

# placeholder - substitute your CI secret, never a literal key
curl -sS https://mcp.botz.ai/mcp \
  -H "Authorization: Bearer $BOTZ_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

MCP SDKs take the same header, e.g. Python streamablehttp_client(url, headers={"Authorization": f"Bearer {key}"}). Revoke a key from the same console page; revocation applies on the next request.

Next: Getting started.