Skip to main content
Glama
CueHQ MCP Server

CueHQ MCP Server

Official
by CueHQSocial

cuehq-mcp

CueHQ finds Reddit, X, YouTube and TikTok threads where people ask for what you sell, ranks them, and drafts replies in your voice. The drafts keep learning from your edits and your own posts. You post every reply yourself. Agents get the ranked threads and the drafts, not raw posts to filter.

This repository is the cuehq-mcp npm package: a small stdio bridge to the CueHQ MCP server at https://www.cuehq.social/mcp, for agents that start MCP servers as a local command. CueHQ never posts and never signs in to your accounts.

Why an agent should use it instead of scraping

  • The threads arrive ranked and deduplicated across four platforms, so the agent reads the ones that matter instead of filtering raw posts.

  • Every thread comes with drafts in your voice, and the drafts keep learning from the edits you make and from what your own accounts post.

  • CueHQ remembers what was answered and what was skipped, so nothing comes back twice.

Related MCP server: @orchyn/mcp

Remote setup

If your agent can add a remote MCP server by URL, you do not need this package.

  • URL: https://www.cuehq.social/mcp (streamable HTTP)

  • Header: Authorization: Bearer cuehq_sk_...

The steps for each agent are at https://www.cuehq.social/agents.

Local setup

  1. Make a key in CueHQ under Settings, Agents. Copy it when it appears, it is shown once.

  2. Add the server to your agent's MCP config. Most stdio clients use this shape:

{
  "mcpServers": {
    "cuehq": {
      "command": "npx",
      "args": ["-y", "cuehq-mcp"],
      "env": {
        "CUEHQ_API_KEY": "cuehq_sk_..."
      }
    }
  }
}

A key acts as you, so keep it out of shared files. Revoke it under Settings, Agents at any time.

Environment:

  • CUEHQ_API_KEY: your key. Every tool call needs it; without it, see below.

  • CUEHQ_MCP_URL: optional. Another endpoint on www.cuehq.social, or a localhost URL for testing.

Without a key the command still starts: it answers initialize and tools/list from a bundled snapshot of the server's tools, and every tool call returns an error that says where a key comes from.

Node 20 or later.

Tools

The bridge adds no tools of its own. These are the seven tools of the CueHQ MCP server:

  • Your brands (brands): the brands on your account, with each one's plan, handles and how many sources it watches.

  • List cues (list_cues): a brand's cues, newest first, by tab (all, priority, mentions, competitors) and by status (surfaced, engaged, skipped).

  • Get a cue (get_cue): one cue in full, with the post, where it is, its top replies and any drafts already written for it.

  • Draft a reply (draft_reply): drafts replies to a cue in your voice, or returns the drafts it already has. Nothing is posted.

  • Mark replied (mark_replied): logs the reply you posted yourself, so CueHQ can track how the thread grows.

  • Skip a cue (skip_cue): skips a cue so it leaves the deck.

  • Weekly report (weekly_report): this week next to last week for one brand.

Pricing

The free plan includes 14 cues. Paid plans start at $19 per brand a month. A key draws on the plan of the account that made it, so the agent sees the same cues and drafts you do.

Privacy and terms

License

MIT. See LICENSE.

Available Tools

7 tools
brandsYour brandsA
Read-onlyIdempotent

The brands on the user's CueHQ account, with each one's plan, handles and how many sources it watches. Call this first when the user has more than one brand: the other tools take its brand_id.

Inputs: no named parameters.

draft_replyDraft a replyA
Idempotent

Drafts replies to a cue in the user's voice, or returns the drafts it already has. Nothing is posted: show the drafts, let the user edit and post the reply themselves, then call mark_replied.

Inputs: id (required).

get_cueGet a cueA
Read-onlyIdempotent

One cue in full: the post, where it is, its top replies and any drafts already written for it.

Inputs: id (required).

list_cuesList cuesA
Read-onlyIdempotent

Cues are live threads where someone is asking for what the user sells. Lists a brand's cues, newest first. scope matches the app's tabs: all, priority (early posts from big accounts), mentions (posts that name the user's handles or brand) and competitors (posts about competitors the user named). status surfaced is the waiting deck, engaged the cues the user replied to, skipped the ones they passed on.

Inputs: limit, scope, status, brand_id.

mark_repliedMark repliedA

Logs that the user posted a reply to a cue, the same as copy + open in the app. Pass the text they actually posted, and draft_id when it started from a draft. CueHQ then checks how the thread grew at 6, 24 and 72 hours. Only call this after the user says they posted.

Inputs: id (required), text, draft_id.

skip_cueSkip a cueB
DestructiveIdempotent

Skips a cue so it leaves the deck, like skip in the app.

Inputs: id (required).

weekly_reportWeekly reportA
Read-onlyIdempotent

This week next to last week for one brand: cues surfaced, replies, exposure (activity on threads after the user replied) and wins, the audience counts on the user's handles, the replies that grew the most, and mentions still unanswered.

Inputs: brand_id.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 7 tools

    Disambiguation4/5

    Most tools map to distinct actions in a clear workflow: discover (brands, list_cues), inspect (get_cue), act (draft_reply, mark_replied, skip_cue), and report (weekly_report). The only mild overlap is list_cues vs get_cue, but one is a filtered collection and the other a single full record, so descriptions disambiguate well.

    Naming Consistency4/5

    Six of seven tools follow a consistent snake_case verb_noun pattern (get_cue, list_cues, draft_reply, mark_replied, skip_cue, weekly_report). The lone bare noun 'brands' breaks the verb-first convention and would more predictably be list_brands, a minor but noticeable deviation.

    Tool Count5/5

    Seven tools is well-scoped for a cue-triage and reply workflow: one entry point, two read tools, three action tools, and one report. Each tool earns its place with no filler.

    Completeness4/5

    Core lifecycle is covered: find brand, list/inspect cues, draft, log the posted reply, skip, and get performance reporting. Minor gaps exist around reversing actions (unskip/restore) or editing an existing draft, but agents can work around these.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Buying-intent judgement for AI agents: three judges decide whether the author of any post or message is a real buyer, and draft a reply you approve.
      106 npm
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables AI assistants to read and analyze real social posts across eight networks, research creators and trends, and generate hooks, draft scores, variants, and repurposed content.
      27
      669 npm
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI assistants to monitor brand mentions across social platforms, assess relevance and buying intent, and generate reply drafts for manual approval.
      MIT