DOCUMENTATION

Hivewire MCP documentation

Connect an agent to Hivewire's hosted news tools. The server works with OAuth-capable clients and agent runtimes that can send an API key.

Hosted MCP server: https://api.hivewire.ai/mcp

Quick start

Add https://api.hivewire.ai/mcp to an OAuth-capable client and sign in with Hivewire. The agent gets 14 tools; you do not need to create or paste an API key.

Choose a client

ENVIRONMENT	CONNECTION	AUTHENTICATION
ChatGPT	Custom MCP server URL	Hivewire OAuth
Codex	Project config + header	Hivewire API key
Claude.ai	Custom connector URL	Hivewire OAuth
Claude Code and other MCP clients	Server URL + header	Hivewire API key
OpenAI Responses API	Remote `mcp` tool	Hivewire API key
Local or open-weight model	MCP-capable agent runtime	Hivewire API key

ChatGPT and Codex

ChatGPT accounts with custom MCP server support can add the Hivewire URL and complete Hivewire sign-in through OAuth. Availability and menu names depend on your ChatGPT plan and workspace settings.

For a project-scoped Codex connection, add this to .codex/config.toml. Replace the placeholder with a key from your Hivewire dashboard:

[mcp_servers.hivewire]
url = "https://api.hivewire.ai/mcp"
http_headers = { Authorization = "Bearer hw_live_..." }

OpenAI documents current Codex configuration fields in its MCP setup guide. ChatGPT does not read the MCP configuration from your Codex host.

Claude

In claude.ai, open Settings → Connectors → Add custom connector and paste the Hivewire URL. Claude opens Hivewire sign-in and consent in the browser.

Claude Code can connect with an API key:

claude mcp add --transport http hivewire https://api.hivewire.ai/mcp \
  --header "Authorization: Bearer hw_live_..."

OpenAI Responses API

Add Hivewire as a remote MCP tool. Keep both the OpenAI and Hivewire keys on your server; see OpenAI's remote MCP documentation for the complete approval flow:

const response = await client.responses.create({
  model: "gpt-5.6",
  tools: [{
    type: "mcp",
    server_label: "hivewire",
    server_url: "https://api.hivewire.ai/mcp",
    headers: {
      Authorization: `Bearer ${process.env.HIVEWIRE_API_KEY}`
    }
  }],
  input: "What are the most significant stories today?"
});

Other clients and open-weight models

The model itself does not connect to MCP. Its agent runtime does. Use a runtime that supports remote MCP over Streamable HTTP and can send an Authorization header. Many clients accept a configuration shaped like this:

{
  "mcpServers": {
    "hivewire": {
      "url": "https://api.hivewire.ai/mcp",
      "headers": { "Authorization": "Bearer hw_live_..." }
    }
  }
}

Authentication is required for every MCP POST, including initialization and tool discovery. Initialization and discovery do not consume credits.

Tools

TOOL	WHAT IT DOES	CREDITS
`search_news`	Semantic / keyword story search; related topics available for pivoting	2
`latest_stories`	The front page, by recency or significance	1
`get_story`	One story in full, source articles attached	1
`related_stories`	Semantically adjacent stories, optionally date-bounded to trace an event	2
`search_articles`	Search individual articles (metadata and links)	2
`trending_topics`	Topics spiking against their baseline	1
`search_topics`	Canonical topic lookup by meaning	2
`build_topic_profile`	Free text → weighted topic profile (LLM)	50
`list_sources`	The publication catalog	1
`list_channels`	Your channels	1
`create_channel`	Create a persistent, learning feed (LLM)	50
`get_channel_stories`	A channel's ranked feed; `since` for deltas	1
`channel_feedback`	Like, dislike, or clear a story to teach the channel	1
`delete_channel`	Permanently delete a channel and free its plan slot	1

delete_channel is irreversible. The tool description warns agents before use. Channel updates are not exposed because an update replaces the learned topic set; edit channels in the dashboard or through the REST API.

Behavior and errors

Errors come back as data. A missing key, exhausted quota, or rate limit appears inside the tool result as JSON with a hint the agent can act on. Rate denials also include retry_after_seconds.

Transport is stateless. The server returns plain JSON over POST. It does not require an SSE stream or a persistent session.

Tool descriptions support exploration. Agents can pivot through a story's topics, trace an event with date-bounded related stories, or follow a subject with a channel that learns from feedback.

Credits

Tool calls use the same credit costs and account limits as their REST equivalents. The MCP connection adds no charge. Your API key and OAuth sessions draw from the same monthly balance.