Hive

npx add-mcp 'https://hivepublish.com/api/mcp'

Connect an assistant to Hive

Hive runs a Model Context Protocol server. Point any MCP client at it and the assistant can publish, schedule, read performance across every connected account, work the inbox, and build DM automations — on the account holder’s own connected channels, under their own consent.

Endpoint https://hivepublish.com/api/mcp

Reviewing Hive for a marketplace or directory?

Hive shows your own connected social accounts, so a brand-new account has nothing in it and every tool answers empty. Email support@hivepublish.com and we will send you a login to a workspace with live channels, published posts, analytics history and running automations, the same day. No card, no trial clock.

Nothing in that workspace can post, reply or message anyone without a separate confirmation inside Hive, so it is safe to exercise every tool.

Connecting

The server speaks Streamable HTTP in stateless JSON response mode, and supports protocol versions 2025-11-25, 2025-06-18, 2025-03-26 and 2024-11-05.

Authentication is OAuth 2.1 with PKCE and dynamic client registration, or a scoped static Bearer API key for server-to-server use. Clients that follow RFC 9728 will find the protected-resource document at /.well-known/oauth-protected-resource/api/mcp; the unauthenticated challenge names it directly. An OpenAPI description of the REST surface is published at /api/v1/openapi.

In a consumer client the whole flow is one step: add the connector, paste the endpoint, sign in to Hive, pick a workspace.

Connecting a specific client

Hive is a remote MCP server: there is nothing to install and nothing to run locally. Clients differ only in whether they can talk to a remote server directly.

Claude

Direct, no bridge. Settings → Connectors → Add custom connector, paste the endpoint, Connect, sign in to Hive, choose the workspace. Available on every Claude plan; the Free plan allows one custom connector. This is the path Hive was first proven on.

Codex, Gemini and other CLI or desktop clients

Clients that expect a local command rather than a URL reach Hive through the standard remote bridge. The config lives in a different file for each client, but the entry has the same shape everywhere:

{
  "mcpServers": {
    "hive": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://hivepublish.com/api/mcp"]
    }
  }
}

The bridge opens Hive’s sign-in page in a browser on first run; after that the client holds the token. If your client takes a remote MCP URL directly, skip the bridge and give it the endpoint.

Your own backend

Issue an API key in Hive under Settings → API and send it as Authorization: Bearer hive_…. Keys are scoped, so a key only carries the permissions you tick when you create it. Use OAuth for anything acting on behalf of another person; use a key only for your own workspace.

What a token can carry

Every tool is gated on a scope, and tools that touch a paid surface check the caller’s workspace role and plan on top of that. Write permissions are off by default: the consent screen shows the read scopes already ticked and every write scope unticked, and the account holder switches on exactly the ones a client should have. They can change or revoke them later under Settings → API. A client cannot request extra permissions for itself — the choice is the account holder’s, every time.

Read

• read:product
• read:channels
• read:calendar
• read:content
• read:analytics
• read:reports
• read:campaigns
• read:inbox
• read:community
• read:listening
• read:automations
• read:podcasts
• read:portfolio
• read:jobs

Write

• write:ideas
• write:drafts
• write:campaigns
• write:reports
• write:schedules
• write:content
• write:community
• write:automations
• write:podcasts
• write:publishing
• admin:workflow

Where a human stays in the loop

Reads and internal drafting run immediately — there is no friction in asking an assistant what happened last week or having it draft ten captions. Anything that reaches a real audience does not work that way.

Eight tools return an approval link instead of doing the thing: scheduling, rescheduling or cancelling a post, replying publicly to a comment, sending a direct message, setting an automation live, and pausing an automation or a flow. There is no immediate-publish tool at all — an assistant cannot put anything in front of an audience on its own.

The workspace owner or an admin confirms the action inside Hive. The approval is exact, single-use and expires after fifteen minutes: it authorises that one call with those exact arguments, and changing the arguments invalidates it. Billing, ownership transfer, provider tokens, workspace deletion and channel disconnection are not exposed to an assistant at all.

Nobody wakes up to something posted in their name that they did not agree to. That is the whole design.

Example prompts

Once Hive is connected, these work in plain language. Each one reaches a different part of the server.

• “Which of my accounts grew the most this month, and what did they post?”
• “Draft five captions for the product launch and check them against Instagram and LinkedIn before you suggest times.”
• “Schedule that carousel to Instagram and TikTok for the best slots next week.”
• “What is sitting unanswered in the inbox, and who has been waiting longest?”
• “Set up a DM automation: when someone comments GUIDE on the launch reel, send them the link.”
• “Add the new podcast episode to my link-in-bio page, above the newsletter.”

The first, fourth and sixth run straight away. The others stop at an approval link, because they would reach an audience.

Tools

47 tools. Reads first, then the ones that change something.

Reading

read_product_knowledge	What Hive can and cannot do, per network
read_channels	Connected accounts and their health
read_calendar	The publishing calendar
read_posts	Published post history
read_analytics_summary	Performance across every connected account
read_best_times	Configured posting slots per channel
read_reports	Saved reports
read_report_schedules	Reports that deliver on a cadence
read_campaigns_and_tags	Campaigns and tagging
read_competitors	Tracked competitor accounts
read_listening	Social listening results
read_inbox_summary	Open, unread and unassigned counts — never message bodies
read_inbox_threads	Redacted conversation metadata, only after the owner opts in
read_reviews	Google Business review work items — never review text
read_automations	Every automation and its state
list_automation_recipes	The six automations an assistant can build, plus the template catalogue
list_connected_accounts	Accounts an automation can run on
read_automation_flow	One automation, step by step
read_automation_results	How an automation actually performed
read_approval_workflows	Who has to approve what
read_content_page	A paginated page of posts, ideas or campaigns
read_advocacy	Employee advocacy groups and feeds
read_podcasts	Podcast shows and episodes
read_portfolio	The workspaces this user may switch between (Queen plan)
read_jobs	Background jobs in flight
read_job_page	A paginated page of background jobs

Writing

create_idea	Add to the ideas library
create_draft	Draft a post
update_draft	Edit a draft
schedule_post	Put a post in the queue — approval required
reschedule_post	Move something already scheduled — approval required
cancel_scheduled_post	Pull something out of the queue — approval required
create_campaign	Start a campaign
create_tag	Add a tag
create_report	Build a report
create_saved_reply	Save a canned reply
reply_to_comment	Answer a comment publicly — approval required
send_direct_message	Send a DM — approval required
create_simple_automation	Build a DM automation
update_simple_automation	Change one
set_automation_live	Turn one on — approval required
pause_automation_flow	Turn one off — approval required
pause_automation	Pause a quick automation — approval required
add_competitor	Track a competitor
queue_competitor_refresh	Refresh competitor data
create_podcast_show	Add a podcast show
cancel_job	Stop a background job

Rate limits

Hive inherits each network’s own limits and enforces them per connected account, so an assistant cannot push an account into a penalty. Instagram and Messenger private replies run at 700 per hour against Meta’s 750 ceiling; DMs at 5,000 per hour smoothed to 100 per minute; public comment replies at 300 per hour; WhatsApp respects the number’s own tier. When a provider rate-limits a call, Hive waits at least as long as the provider asks before retrying. When a provider restricts an account outright, Hive stops sending on it and leaves it stopped until a workspace admin reviews and resumes it in Hive — it never quietly resumes itself.

Getting access

You need a Hive account with at least one connected channel. Every plan, including the 14-day trial, can use the connector — there is no plan gate on access itself. Plans differ in monthly request volume: 7,500 on Worker Bee, 15,000 on Swarm, 50,000 on Queen Bee. A few tools are plan-specific and say so when called. API keys are issued in Hive under Settings → API.

Questions: support@hivepublish.com.