Agentic tools · MCP

Put an AI agent on your reviews.

Reply Argus is a remote MCP server, so any AI agent can run your review desk (triage, analytics, draft and publish) over a secure login scoped to your workspace. Add it in one click, or wire the raw endpoint yourself. Pick your path below.

MCP URLhttps://www.replyargus.com/api/mcp
Pick your way inSee the tools

Add it to your setup

One action, per client.

Cursor and Claude add in a click; Claude Code and Codex are a copy-paste. Every path signs you in over OAuth, never an API key.

Cursor

1-click

One click. Cursor runs the OAuth sign-in for you.

Claude Code

Copy & run

Copies the CLI command. Run it, then `/mcp` in a session to sign in.

Claude Desktop

1-click

Opens Claude with the connector prefilled. Review, confirm, sign in.

Codex

Copy & run

Copies the config for ~/.codex/config.toml. Then run `codex mcp login replyargus`.

OAuth, not API keys

Sign in with your Reply Argus account. Revoke anytime from settings.

Scoped to your workspace

Every call runs as you. The agent only ever sees your organization's data.

Approval by default

Drafting never posts. Publishing to the stores always asks you first.

Ways in

Four ways to turn it on.

Same tools underneath. Pick the path that fits how you work. Each takes a minute.

Claude connector

One-click

For: Anyone who works in Claude, the fastest way in.

Add one URL and every tool below appears inside Claude, over a scoped OAuth login.

  1. 1

    Add the connector

    In Claude → Settings → Connectors → Add custom connector, paste the MCP URL.

    https://www.replyargus.com/api/mcp
  2. 2

    Sign in once

    Claude sends you to Reply Argus to approve access. Everything stays scoped to your workspace.

  3. 3

    Just ask

    “Draft replies for my new 1-star reviews.” Claude calls the tools and shows you the drafts. Try the triage_inbox prompt.

Raw MCP endpoint & OAuth

Developers

For: Developers wiring their own MCP-capable client or agent.

A standard remote MCP server (Streamable HTTP) behind OAuth 2.1. Point any MCP client at it.

  1. 1

    Point your client at the server

    It speaks the Model Context Protocol over Streamable HTTP at:

    https://www.replyargus.com/api/mcp
  2. 2

    Authorize over OAuth 2.1

    The server advertises its authorization metadata; your client runs the standard OAuth flow (authorization-code + PKCE) and receives a bearer token scoped to the user's workspace. Every tool call runs under that identity. RLS enforces isolation server-side.

  3. 3

    Call tools, read resources, use prompts

    The full tool/resource/prompt catalog below is discoverable over the protocol (tools/list, resources/list, prompts/list). No extra setup, the same surface the Claude connector uses.

Interactive reply card

No-code

For: People who want to click, not read.

An editable draft→publish card renders right inside the chat. Edit the reply and hit Publish, no retyping.

  1. 1

    Add the Claude connector

    Nothing else to install. The card rides on the connector you already added.

  2. 2

    Open the card

    Ask Claude to “open the reply card” for a review. It appears as an interactive card in hosts that support MCP Apps; edit and Publish inline.

Reply-craft Agent Skill

Power users

For: Power users who want expert-quality, on-brand replies.

A reply-craft playbook (tone per language, the 1-star framework, escalation) that drives the tools so Claude replies like a pro.

  1. 1

    Add the Claude connector first

    The skill drives the connector's tools, so it needs to be connected.

  2. 2

    Install the skill

    claude.ai / Desktop: Settings → Capabilities → Skills → upload the reply-argus-review-craft folder. Claude Code: copy it into ~/.claude/skills/.

  3. 3

    Let it work

    It auto-triggers on review-reply work, or invoke it directly with /reply-argus-review-craft.

The toolkit

18 tools, four jobs.

Everything the dashboard does, callable in plain language. Read tools just look; write tools change things; Confirms tools post publicly and always check with you first.

Read & search

Pull up any review, filtered any way, without leaving the chat.

search_reviewsRead

Filter by app, rating, status, sentiment, theme, language, text, or date, paginated.

list_reviewsRead

Recent reviews for an app, newest first, with the current reply.

get_reviewRead

One review in full, with its reply history.

list_appsRead

Every app in your workspace and its store-connection health.

list_knowledgeRead

What Argus knows about an app (store listings it learned per platform, your marketing page, and manual notes) plus any contradictions it flagged between them.

usage_and_limitsRead

Your plan's reply/app quota: used, remaining, and reset.

whoamiRead

The signed-in account, organization, and plan.

Insights

The same analytics as your dashboard, asked in plain language.

analytics_overviewRead

Volume, average rating, sentiment split, top themes, and reply status over a window.

theme_breakdownRead

The recurring themes driving your reviews, with counts and context.

weekly_reportRead

The AI executive digest for the week: narrative plus the numbers behind it.

Draft & publish

Write on-brand replies and post them, with your approval by default.

draft_replyWrite

Generate an on-brand reply grounded in your persona, everything Argus learned about the app (store listing, marketing page, your notes), and past approvals. Returns text, never posts.

translate_reviewWrite

Translate a review into your language before you respond.

open_reply_cardConfirms

Open an interactive card for a review. Edit the draft and hit Publish inline, in hosts that support MCP Apps. Posts via the same human-approved path.

publish_replyConfirms

Post a reply to the App Store / Google Play. Public and permanent. Claude confirms first.

Automate & manage

Run the watch: sync, cadence, and autonomous publishing.

sync_reviewsWrite

Pull the latest reviews from the stores for an app now.

toggle_agentWrite

Turn review monitoring on or off for an app.

set_draft_cadenceWrite

How often the agent auto-drafts: off, daily, or hourly.

set_auto_publishConfirms

Enable or disable autonomous auto-publishing for an app, changes what posts without you.

One-tap workflows

Prompts

Pre-built flows that chain the tools. Pick one and the agent runs the whole play.

  • triage_inbox

    Surface and prioritize unanswered negative reviews, then propose drafts.

  • weekly_review_report

    Generate the weekly digest and KPIs for an app.

  • draft_and_review

    Draft a reply for one review and let you approve before it posts.

Live context

Resources

Reference live data with @, no tool call needed.

  • review://{id}

    A single review, referenceable as live context.

  • reviews://{appId}

    Recent reviews for an app.

  • digest://{appId}

    The latest digest for an app.

  • knowledge://{appId}

    What Argus learned about an app: store listings, marketing page, your sources, and open contradictions.

From the blog

How the agent works

Is It Safe to Auto-Publish Replies to App Reviews? The 2026 Guardrail Guide
The MCP Server for App Reviews: Connect Your AI Agent to Your Store Reviews
Manage Your App Store Reviews From Claude: The 5-Minute MCP Setup
Using ChatGPT to Respond to App Reviews: The Prompt Pack and the Refund Trap

Agentic FAQ

AI replies, without the risk.

How the agent drafts, what it can touch, and where you stay in control.

Explore the toolkit

Is it safe to let an AI reply to my reviews?

Yes, because a human stays in the loop by default. Drafting never posts anything, and publishing to a store always asks you first unless you have explicitly turned on auto-publish for a scope you chose. Every draft is grounded in facts about your app and obeys your brand rules, so the agent triages, drafts, and analyzes freely while public replies stay under your control.

Does a human still approve what gets published?

Yes. Approval is the default: the agent can draft, translate, triage, and pull analytics on its own, but publishing a reply to the store routes through the same approval gate as the dashboard. Auto-publish is opt-in and off until you enable it, and even then only for the replies you scope (like 5-star thank-yous).

What is the MCP connector and how do I add it?

Reply Argus is a remote MCP server, so any AI client can run your review desk (triage, analytics, drafting, and publishing with approval) over a secure login scoped to your workspace. You add it by URL in Claude, Claude Desktop, Claude Code, Cursor, Codex, or any MCP client. It signs you in over OAuth rather than an API key, and you can revoke access anytime. The connector is available to workspace owners and admins.

Will the AI make things up in a reply?

Argus is built to avoid that. Every draft is grounded in your approved replies and the facts it reads from your store listings and marketing page, and when you open a draft it shows the exact notes and listing lines it drew from, quoted as you wrote them. If two sources disagree it flags the mismatch, and your brand facts (support email, refund link, the lines you never want said) are obeyed on every reply. Mark a cited fact right or wrong and the next draft leans on what you confirm.

Do the app stores allow replies written with AI?

Both Apple and Google let developers respond to their own reviews, and Reply Argus posts through those official developer-response channels using your own connected developer account, exactly as if you typed the reply yourself. It never impersonates a user or posts outside the sanctioned reply flow, and approval-by-default keeps you as the author of record for anything that goes live.

Do I need to write code or manage API keys to use it?

No. Every path signs you in over OAuth, not an API key. In Cursor and Claude it adds in a click; in Claude Code and Codex it is a copy-paste of the endpoint. Once connected, you drive the whole review desk in plain language, and the same tools work identically underneath whichever client you pick.

Security & data

Built to be trusted with your data

The protections behind every review we read and every reply we draft.

Encrypted, and yours alone

Your store credentials are encrypted at rest and in transit. We never see your raw keys.

Isolated per workspace

Reviews, replies, and knowledge are isolated at the database layer. One workspace can never read another's.

We never store card data

Payments run through a merchant of record. Card details never touch our servers.

Scoped, revocable access

Role-based permissions, OAuth for integrations, and hashed API keys you can revoke anytime.

Nothing ships without your say-so

Drafts never publish on their own. Auto-publish is an explicit opt-in, off by default.

Hardened by default

Strict security headers, a content security policy, enforced HTTPS, and signed, verified webhooks.

How we protect your data

Give your reviews an agent.

Connect your workspace and let AI triage, draft, and publish (with you in the loop).

Start free
See all features

Add by URL in Claude, Claude Desktop, Claude Code, Cursor, Codex, or any MCP client. Revoke access anytime.