ProductClank MCP Server
Launch community growth campaigns from your AI assistant. ProductClank turns real communities (and their agents) into a distribution network — this MCP server lets Claude, ChatGPT, or any MCP client rally that network on your behalf: boost a post with real replies, likes, and reposts; launch community content campaigns; draft social content into your own pipeline.
- Remote server:
https://mcp.productclank.com/mcp(Streamable HTTP) — see Endpoints for the creator-campaign endpoint - Auth: OAuth 2.1 — sign in with your ProductClank account, revoke anytime
- Billing: your ProductClank credits, per-tool costs shown below, with per-app daily spend caps
- Website: productclank.com/mcp
The same URL works in both Claude and ChatGPT — the only difference is where each app hides its "add a connector" screen (and, for ChatGPT, a plan/Developer-mode requirement). Once connected, both drive the identical tools below.
Endpoints
The server mounts the same MCP implementation at two URLs, each serving a
different tool profile (src/tools/index.ts):
| Endpoint | Profile | Served to | Tools |
|---|---|---|---|
https://mcp.productclank.com/mcp | full | Claude, ChatGPT, the MCP registry — the URL in every connect guide below | all of them, boost_post included |
https://mcp.productclank.com/creator | creator | Meta's Muse connector directory | everything except boost_post |
The creator profile is the connector as a creator-campaign product: a brand
commissions creators, and creators are paid for original content they publish
from their own accounts. boost_post is the one tool that does not fit — it
sells likes and repost, which are engagement signals, and accepts Instagram
post URLs — and Meta's Spam community standard bars paying for engagement. So
it is excluded from /creator pending Meta approval, after which it can be
added back.
The split is by URL rather than by OAuth client on purpose: tools/list is
deliberately unauthenticated (so directory health checks can introspect), which
means a reviewer can list tools before any client identity exists. Prompts and
the productclank://capabilities resource are filtered the same way, so nothing
the creator endpoint serves mentions boosts, likes or reposts.
Both endpoints share one OAuth authorization server, one set of tokens, one
session sweeper and the same JSON-only-client handling; each is its own RFC 9728
protected resource (/.well-known/oauth-protected-resource/mcp and
…/creator).
Connect from Claude
- In Claude, go to Settings → Connectors → Add custom connector (Desktop or web).
- Paste the URL:
https://mcp.productclank.com/mcp - Approve the OAuth prompt — it signs you into ProductClank (Google or email) and asks for consent.
- Ask Claude things like:
"Boost this post with replies from the ProductClank community: https://x.com/…" "Preview a content campaign for my product." "What's my ProductClank credit balance?"
Getting new tools: Claude loads the tool list when a conversation starts.
After we ship a new tool or option, start a new conversation to see it. In
Claude Code, run /mcp and reconnect ProductClank (or start a new session).
Connect from ChatGPT
ChatGPT reaches custom MCP connectors through Developer mode.
Requirements — Custom connectors need a paid ChatGPT plan (Pro, Plus, Business, Enterprise, or Education) on the web app. They are not available on the free tier, so free-tier users won't see an "add connector" option — this is a ChatGPT limitation, not a ProductClank one.
- In ChatGPT on the web, open Settings → Apps & Connectors → Advanced settings and turn on Developer mode.
- Back in Apps & Connectors, click Create / Add custom connector.
- Give it a name (e.g.
ProductClank), a short description, and the MCP server URL:https://mcp.productclank.com/mcp - Choose OAuth as the authentication method and connect — it signs you into ProductClank (Google or email) and asks for consent, exactly like Claude.
- In a chat, enable the ProductClank connector and ask it to boost a post, preview a campaign, or check your balance.
Works in any MCP client that supports remote servers with OAuth (Claude web/desktop, Claude Code, ChatGPT Developer mode, and others).
Tools
Grow — run growth campaigns for a product
| Tool | What it does | Cost |
|---|---|---|
search_products | Resolve your ProductClank products to a product_id | free |
create_product | List a product from just its website URL (token-free listing, auto-filled) | free |
create_campaign | Create a discovery campaign: continuously find relevant posts + draft community replies | 10 cr |
run_research / get_research | Topic & competitor analysis — the free pre-flight before spending | free |
generate_posts | Run discovery: scrape matching posts and draft a reply for each | 12 cr/post |
list_campaigns / get_campaign / get_posts | Campaign inventory, config + stats, discovered posts & drafts | free |
review_posts | AI-score posts against relevancy rules, prune the irrelevant (dry-run billed too) | 2 cr/post |
regenerate_replies | Redraft selected replies with an edit request | 5 cr/reply |
get_campaign_activity | What's new since the last check — posts found, replies claimed, live links to what went out | free |
get_campaign_results | Spend vs outcomes: the funnel, approval + survival rates, cost per usable reply | free |
update_campaign | Merge keywords, enable the discovery sources research found, move the relevance bar, pause/resume, re-aim the platform targeting, change who posts the drafts | free |
set_campaign_schedule | Put discovery on a schedule so it keeps finding conversations between sessions. Shows the projected daily cost and requires an explicit yes before enabling | free to set · 12cr per post found |
add_delegate | Hand a campaign to a human to manage in the web app | free |
boost_post (/mcp only — see Endpoints) | Rally the community to engage a specific post — AI-drafted replies or quote posts at 20 cr per draft (count 1–10, default 10 = 200 cr; quote posts X only — a repost with each member's own text, landing in their followers' feeds), 30 likes, 10 bookmarks (X only) or 10 reposts (300 cr). Auto-detects platform from the URL: X, Instagram, TikTok, LinkedIn, Reddit, Farcaster, YouTube | 20 cr/draft or 300 cr |
Earn — participate in campaigns for the connected user
| Tool | What it does | Cost |
|---|---|---|
find_opportunities | Browse unclaimed reply and quote-post drafts the user can post to earn (likes and reposts need a screenshot and stay in the web app) | free |
submit_participation | Submit the posted reply's URL (X, Reddit, YouTube, LinkedIn) — attributed to the user's linked handle for that platform, then points/credits are awarded | earns |
find_open_campaigns | Discover content & take-action campaigns the user can join — public ones plus their communities' | free |
get_campaign_brief | Read a campaign's full brief: what to create or do, judging criteria, rewards, deadline, remaining allowance | free |
submit_campaign_work | Submit the user's work: a content or action-proof URL, a hosted image/video made for the task (media_url — links only, no uploads), and/or a description. X posts are author-verified. Pending → owner review → Stars/points | earns |
get_my_submissions | Track submission status + the reviewer's notes | free |
get_earnings | The user's participation totals: points, credits, reply stats | free |
Content Studio
| Tool | What it does | Cost |
|---|---|---|
suggest_content_campaign | AI-drafted preview of a community content campaign (title, description, CTA) + affordability check. Nothing is created | free |
create_content_campaign | Launch the content campaign: the community creates posts/threads/videos for your product; submissions and winner selection happen in the web app | 1,000 cr |
list_content_spaces | List the content spaces you can draft into | free |
get_content_workspace | The brand's calibration — voice, platforms, post types, playbook, topic inventory, style guides. Read it before drafting | free |
setup_content_space | Turn content on for a brand from an onboarding chat (pair with the setup_content_space prompt): voice, platforms, post types, topics — can create a solo space for the brand. Also updates settings later | free by fields · 5 cr for a pasted brand doc |
manage_content_topics | The inventory of aspects the brand talks about: list / add / update / remove / suggest | free |
write_content_candidates | Draft up to 25 posts into the space, in the calibrated voice. They are auto-scored by the reviewer and wait in the queue for the user's approval; nothing auto-publishes | free |
get_content_queue | The queue as the user sees it: each draft with its score, verdict, and the reviewer's one-line fix | free |
revise_content_draft | Approve (stage), drop (discard), edit, or nudge a draft with a one-click preset — shorter, longer, punchier, deeper, simpler, more_specific, less_salesy, more_casual, more_formal — or a typed note; fix / humanize / review | stage · discard · edit free; rewrites & review 2 cr |
teach_content_voice | Turn a reaction into a standing rule in the brand's voice KB so every future draft honors it | 1 cr |
Credits
| Tool | What it does | Cost |
|---|---|---|
check_balance | Your credit balance and plan | free |
credit_history | Your credit transactions (spend + rewards), newest first | free |
report_issue | Report a failing tool or a dead end to the ProductClank team (a human replies) | free |
get_support_status | Your reported issues and the team's replies | free |
The server also exposes two MCP prompts — grow_product, a ready-made operating
procedure for the growth loop, and setup_content_space, the Content Studio onboarding
interview (brand → audience → tone → example posts → platforms → post types → topics,
then first drafts and the approve / tweak / teach loop) — and a resource —
productclank://capabilities, the tool/cost roster — so agents can plan spend before
calling anything.
Costly actions are designed to be confirmed with the user first — tool descriptions instruct the model to preview and state the credit cost before spending.
How auth & billing work
The server is an OAuth 2.1 authorization server backed by the ProductClank web app (/connect/mcp login + consent). Every /mcp request requires a valid access token; tools act as the connected user and bill that user's credit balance. Users can see and revoke the connector — and set a daily credit spend cap per connected app — from their ProductClank profile ("Connected Apps").
Communication contract
Tool results are read by an assistant, not by the person paying the credits, so anything the user must know or decide travels in the payload:
user_note— one or two plain sentences for the assistant to relay.decision_offer— a real choice (question+optionswithwhat_happensandcost+current+how_to_apply) for the assistant to present, never to answer on the user's behalf.
The main one is distribution: campaigns are created private (drafts wait in
the owner's workbench) and the result offers community distribution — the
reach the platform exists for — as an explicit, costed choice. Spends get the
same treatment: suggest_content_campaign returns the 1000-credit launch as a
yes/no, and a review_posts dry run says what it already cost and what the
next step actually changes.
Telemetry
The server emits three PostHog events — mcp_connected (OAuth consent
completed), mcp_tool_called (every tool, with ok and duration_ms) and
mcp_tool_error (failures, with the message) — so the connect → first-call →
repeat-use funnel is measurable and refused calls double as roadmap signal.
Events are keyed to the user's Supabase auth id where resolvable, so connector
activity joins the same PostHog person as their webapp activity. Capture is
fire-and-forget and the whole thing no-ops without POSTHOG_API_KEY. See
src/lib/analytics.ts; tools are instrumented
centrally in src/tools/instrument.ts.
Architecture
A thin Express + @modelcontextprotocol/sdk wrapper over ProductClank's public agent REST API (/api/v1/agents/*). The REST API is canonical; this server and the ProductClank agent skill are parallel wrappers — capabilities.json is the parity source of truth and CI fails when they drift (npm run check:parity).
Claude / MCP client ──▶ mcp.productclank.com/mcp (profile: full)
Meta Muse directory ──▶ mcp.productclank.com/creator (profile: creator, no boost_post)
│ OAuth 2.1 AS ──▶ productclank.com/connect/mcp (login + consent)
└─▶ ProductClank agent REST API (/api/v1/agents/*)
Development
cp .env.example .env # fill in (see DEPLOYMENT.md)
npm install
npm run dev # tsx watch
npm run typecheck
npm run build && npm start
See DEPLOYMENT.md for production deployment (Docker / Railway) and CAPABILITIES.md for the tool roadmap.
Operations
When a release depends on new app-repo behavior, deploy the app repo first; tools must degrade gracefully against the older API.
Registry
Published to the official MCP Registry as com.productclank/productclank — see server.json.
Links
- What is ProductClank? · MCP connector page · For agents & developers
- Issues & support: GitHub issues or in-app support