Use in your agent

Wire buzzabout's MCP server into any MCP-capable client — Claude, Claude Code, Codex, Cursor, ChatGPT, or your own SDK-built agent.

The buzzabout MCP server speaks the Model Context Protocol — any MCP-capable client can connect. This page covers the wiring for the common hosts plus the SDK pattern for your own agent.

Endpoint

https://mcp.buzzabout.ai/mcp/

Streamable HTTP. Trailing slash required. Standard assistants (Claude Desktop/web, ChatGPT) authenticate via OAuth; CLI / IDE agents (Claude Code, Codex, Cursor) and custom agents use an x-api-key header — see Authentication.

Wire it up

Settings → Connectors → Add custom connector (the +). Fill in:

Field	Value
Name	`buzzabout`
URL	`https://mcp.buzzabout.ai/mcp/`

The connector is added under Not connected. Click Connect to sign in with your buzzabout account (OAuth) — it stays disconnected and unusable until you do. Once connected, Claude replays the token on later calls. Identical on Claude Desktop and Claude.ai (web).

Claude Code uses an API key (no OAuth flow for custom MCP servers). Mint a key under Settings → API keys in the buzzabout web app, then:

claude mcp add --transport http buzzabout \
  https://mcp.buzzabout.ai/mcp/ \
  --header "x-api-key: bz_live_..."

Check it's connected with claude mcp list.

Add an entry to ~/.codex/config.toml. Codex authenticates a streamable-HTTP server with static headers — use x-api-key (mint a key under Settings → API keys in the buzzabout web app):

~/.codex/config.toml

[mcp_servers.buzzabout]
url = "https://mcp.buzzabout.ai/mcp/"
http_headers = { "x-api-key" = "bz_live_..." }

Settings → Cursor Settings → MCP → Add new MCP server.

~/.cursor/mcp.json

{
  "mcpServers": {
    "buzzabout": {
      "transport": "streamable-http",
      "url": "https://mcp.buzzabout.ai/mcp/",
      "headers": {
        "x-api-key": "bz_live_..."
      }
    }
  }
}

Cursor doesn't ship an OAuth flow for custom MCP servers — use the x-api-key path. Mint a key under Settings → API keys in the buzzabout web app.

Settings → Apps → Create app, enter the server URL, and select OAuth as the auth mechanism. Requires a paid ChatGPT plan (Plus, Pro, or Team).

Field	Value
URL	`https://mcp.buzzabout.ai/mcp/`
Auth	OAuth

ChatGPT then prompts you to sign in with your buzzabout account.

For your own agent, use one of the official MCP SDKs with an x-api-key header.

agent.py

import asyncio
import os

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client


async def main() -> None:
    headers = {"x-api-key": os.environ["BUZZABOUT_KEY"]}

    async with streamablehttp_client(
        "https://mcp.buzzabout.ai/mcp/",
        headers=headers,
    ) as (read, write, _):
        async with ClientSession(read, write) as session:
            await session.initialize()

            tools = await session.list_tools()
            print(f"loaded {len(tools.tools)} buzzabout tools")

            result = await session.call_tool(
                "buzzabout__list_datasets",
                arguments={"limit": 5},
            )
            print(result.content)


asyncio.run(main())

A first prompt

Once the server is wired in, try this in any MCP-capable assistant:

"Research how people talk about cold brew on Reddit over the last week, then show me the top themes and a few representative posts."

The assistant routes the whole thing through buzzabout__ask — it writes the query, previews it, collects posts, and analyses — then polls for the answer:

buzzabout__ask("research cold brew on Reddit, last week …") → returns { chat_id, message_id, status: "working" } immediately.
buzzabout__get_message(chat_id, message_id) — long-polls until stop_reason is non-null (the run can take a few minutes).
Walk the returned blocks: a render: true block → buzzabout__render(message_id, block_id) shows it as an interactive widget; a render: false block carries its text to relay.
Continue the thread by passing the chat_id back into ask.

Every tool call is visible in the host's UI — inspect what was sent and what came back.

No create/update/delete tools

Collecting datasets, profiling audiences, and detecting patterns all happen through ask — there are no direct CRUD tools. The other tools are read-only lookups. See the tools reference.

Troubleshooting

Empty tool list or silent failure

Trailing slash? https://mcp.buzzabout.ai/mcp/ (with trailing slash). The unslashed /mcp redirects via 307, which strips the request body in many MCP clients.

Auth? OAuth tokens expire — sign out and back in via the host's integration settings. API keys can be revoked via the buzzabout web app's Settings → API keys.

Plan? MCP access is gated by plan tier — your plan must include MCP. GET /v1/me reports whether it's enabled.

If the tools list is empty, the most likely causes are:

Auth failure — sign out and back in (OAuth) or check key validity (x-api-key).
Network proxy / firewall — outbound HTTPS to mcp.buzzabout.ai/mcp/ is blocked.
Account lacks access — your plan must include MCP.

Still stuck? Mail support@buzzabout.ai with the request id from any failed call.

Next steps

Tools reference — every tool the surface exposes.
Authentication — which auth method per client.
MCP overview — the surface at a glance.

Endpoint

Wire it up

A first prompt

Troubleshooting

Next steps

On this page