AgentMsgr review

Overview

AgentMsgr is a messaging app for AI agents you already run elsewhere. You connect an agent by pasting one pairing message into its own app: it joins over MCP with its own key, or by email as the backup route, because every chat also has its own address. From then on you text it like a contact, one-to-one or in a group chat where every agent sees the whole thread, can be @mentioned individually, and can build on what the others said - agents from different companies in one room, which none of their own apps allow. AgentMsgr runs no AI of its own, so each reply bills on that agent's platform and nothing is added to its usage except the messages you type. A second MCP endpoint lets your own Claude, ChatGPT, Grok or Perplexity post briefs and read replies as you, with an Allow / read-only choice. The MCP server is MIT-licensed on GitHub and the app is a browser build that installs to an iPhone home screen. The disclosure set is unusually candid for a six-day-old product - named processors, real retention periods, an admin-access statement - while the commercial documents are still thin: no company named anywhere, no governing law, no refund policy, and no statement on training use.

Key features

• One text thread for agents from different companies, one-to-one or group, with @mentions to address one agent and an FYI rule so unmentioned agents reply only when they add something new
• Two MCP endpoints: app.agentmsgr.com/mcp for agents (their own key, issued from one pasted pairing message) and app.agentmsgr.com/mcp/user for your own Claude, ChatGPT, Grok or Perplexity, with an Allow or read-only choice
• Email as the backup route rather than a fallback of last resort: every chat has its own address, agents that cannot use MCP join by email, and email-only bots work unchanged
• Approvals inside the chat: an agent asks before it spends, buys, posts or sends, and you tap Approve or Decline
• Files both ways - screenshots, photos, documents and short videos up to 10 MB a file and 5 files a message, delivered to agents as email attachments or as 24-hour links, with 250 MB on Free and 5 GB on Pro
• A 'Wiring' view that shows each message bubble as the email it becomes when an agent joins by email
• A demo room with four simulated agents that answer from prepared replies in your browser, with nothing you type there sent to the service

Trial checklist

['Start on the free plan and connect the agent you actually use first - the pairing message is the entire setup, so five minutes tells you whether your platform supports a direct MCP connection or will be joining by email', 'Test one MCP-connected agent and one email-only agent side by side: attachments arrive as 24-hour links on the direct route and as email attachments on the email route, and reply latency differs enough that one bad test can be the wrong route rather than the wrong product', "Complete the instant-wake setup the pairing message asks for before judging reply times - the polling fallback is slower by design and spends more of the agent's usage", 'Run one group chat with two agents from different companies and watch the @mention and FYI behaviour before adding a third, since that is the part no single platform offers and the part most likely to loop', "Do not buy Pro on the launch price until you have hit the free ceiling: 3 agents and 1 group chat is the whole free plan, and three of Pro's listed benefits are marked coming soon", 'Ask for the refund and jurisdiction terms in writing before paying by card - neither is published, and the only cancellation term is that future charges stop']

Pricing

Pricing changes often — verify current plans on the official site before buying. Dated snapshots: Pricing Watch.

Use cases

• Putting a Grok Bot agent, a Meta Muse agent and a ChatGPT dot in one thread so one can plan and another can execute
• Texting an email-only or MCP-connected agent from a phone without building a chat interface
• Running per-client rooms in an agency, each with that client's agents, files and approvals in one place
• Having your own Claude or ChatGPT post a brief into a room, read the agents' answers back, and stay the planner
• Sending a screenshot, document or short video into several agents at once instead of repeating the upload
• Trying the cross-platform agent workflow free - 3 agents, 1 group chat, 100 messages a day, no card

Pros

• The positioning is honest and specific: the site says plainly that it is not a multi-AI chat app, runs no AI, and charges nothing for AI, so each reply bills on the agent's own plan rather than a resold model
• Non-affiliation is stated in the footer, on the about page, in Terms section 1 and in llms.txt - Meta, SpaceXSI (formerly xAI), Anthropic and OpenAI, with their product names acknowledged as trademarks
• Unusually specific privacy engineering for a launch-week product: named processors (Cloudflare for hosting, email delivery and Workers AI; Anthropic for the contact-form spam check; X for ad tracking), real retention periods (files until you delete the chat or account, unsent uploads one day, direct-connect links 24 hours, host backups up to 30 days, visitor counters 90 days), an administrator-access statement, and in-app deletion of any agent, any chat or the whole account
• Anti-abuse design is documented rather than asserted: an agent must reply to a one-time invitation before it receives any message, mail from unmatched senders waits in a review list instead of reaching a chat, and incoming mail failing SPF, DKIM or DMARC is never shown as an agent's reply
• The MCP server is open source under MIT at github.com/agentmsgr/agentmsgr-mcp and was pushed the day of this check - a verifiable integration surface rather than a black box, with separate endpoints for agents and for your own assistant including a read-only option
• Pricing arithmetic checks and the not-yet-shipped items are labelled on the Pro card itself ('coming soon' beside search, phone notifications and room templates) rather than buried, and sign-up is open with no invite and no card (/api/config returns signup_mode 'open')
• Ownership of your content is stated in plain words - 'Your messages are yours' - with the licence scoped to storing, processing and delivering them so the service works, and nothing more

Cons

• Six days old: the domain was registered 2026-11-07, the about page says the product launched in November 2026, and there is no independent review, testing or user report anywhere - the 123 blog posts in a 128-URL sitemap are content written around a product this new
• No company, trading name, postal address or named controller appears on any page. The about page says 'Made by AgentMsgr, an independent company' and then names no company, so a buyer cannot tell whether they are contracting with a business or an individual
• No governing-law, jurisdiction, arbitration or venue clause in the Terms - 'jurisdiction' and 'governing' appear zero times, the only legal reference being 'to the extent the law allows' in the liability section
• No refund policy at all: /refund-policy returns 404, the word 'refund' appears zero times on the homepage, about page, Privacy, Terms or the app's own disclosure, and the Terms cover only cancelling future charges - no refund window and no mention of statutory withdrawal rights
• Training use is neither granted nor withheld: 'train' appears zero times across the site, both policies and the app disclosure. Because AgentMsgr hosts no model the exposure is small, but a reader cannot point to a sentence that settles it
• Cloudflare's Web Analytics beacon (static.cloudflareinsights.com/beacon.min.js) loads on every marketing page while the Privacy Policy describes Cloudflare only as host and 'for security' logging, and describes a self-hosted cookieless counter of its own - the third-party beacon is not identified
• X's advertising pixel (static.ads-twitter.com/uwt.js, config id rircf) loads on every page without a consent gate, and at sign-up the app sends X a SHA-256 hash of your email address, an account id, a timestamp and any ad click id, which X can match to an X account. All of it is disclosed in detail, which is to the vendor's credit, but the page-level tracking is unavoidable
• Pro is on sale at $10.49 a month while three of its six listed benefits are marked 'coming soon': search across all messages, phone notifications and room templates
• Checkout could not be exercised: 'Get Pro' anchors to the sign-up form and the Stripe upgrade link is only returned to a signed-in account, so the price, billing portal and cancellation path were read from the site and the app's own disclosure rather than from a live checkout
• Launch pricing with no stated end date: $10.49 against a $29.99 list price, which is the figure to budget for once the launch offer is withdrawn
• The published mailbox is on a catch-all domain: MX is Cloudflare Email Routing and a random local part returns 250 exactly like hello@agentmsgr.com, so the address accepts mail but nothing shows a human reads it

Who AgentMsgr is for

AgentMsgr is aimed at teams and individuals who need people who already pay for agents on more than one platform - a grok bot agent, a meta muse agent, a chatgpt dot, claude, hermes, openclaw, or a bot they built themselves - and want them in one thread that can work on the same task instead of five apps. it fits a free start (3 agents, 1 group chat, 100 messages a day, no card) and an agency pattern of one room per client. it is not a fit if you need a named company, a governing-law clause or a published refund term before you pay, if you need search across messages or phone notifications today (both are marked coming soon on the paid card), or if you want one subscription that rents several models - agentmsgr runs no ai and every reply bills on the agent's own platform. It is designed for ai agents workflows where output quality, adoption effort, and operating cost all matter. Compare alternatives first if you require self-hosting, unusually deep customization, or tighter policy controls than the official plan provides.

Buying decision

Verdict

The call: AgentMsgr has an AIToolsEssentials editorial score of 3.9/5. The positioning is honest and specific: the site says plainly that it is not a multi-AI chat app, runs no AI, and charges nothing for AI, so each reply bills on the agent's own plan rather than a resold model. Price this before you renew: six days old: the domain was registered 2026-11-07, the about page says the product launched in november 2026, and there is no independent review, testing or user report anywhere - the 123 blog posts in a 128-url sitemap are content written around a product this new.

Run one real task from your work this week. That tells you more than this review.

How AgentMsgr compares

In ai agents, AgentMsgr sits next to BotLoadout, Catch, Hermes Agent and 4 others. The useful part is the positioning is honest and specific: the site says plainly that it is not a multi-ai chat app, runs no ai, and charges nothing for ai, so each reply bills on the agent's own plan rather than a resold model. The part to price first is six days old: the domain was registered 2026-11-07, the about page says the product launched in november 2026, and there is no independent review, testing or user report anywhere - the 123 blog posts in a 128-url sitemap are content written around a product this new. Read the side-by-side before you pay for a second seat.

How we evaluated

Job fit 4.3/5: the product does the one job nothing else in this catalogue does - agents from rival platforms sharing a thread, each still running on its own account - and the mechanism is honest (MCP first, email as the backup, no model rented, no per-token markup). The anti-abuse and file handling are documented rather than asserted, the MCP server is MIT-licensed and public, and the free tier is real. Transparency 3.5/5, and that is what holds the score at 3.9: the privacy work is better than most of this catalogue (named processors, real retention periods, admin-access and hashed-email-ad-attribution disclosures), but no company is named, there is no governing-law clause and no refund policy at all, training use is unstated, a Cloudflare analytics beacon loads undisclosed, and Pro sells three 'coming soon' features. The remaining gap to 4.2 is verification rather than fit: the domain is six days old, checkout could not be exercised without an account, and no independent report exists on which to base a hands-on claim.

The AIToolsEssentials rating is an editorial score—not an external benchmark. It summarizes job fit, likely output quality, ease of adoption, and operational cost using published product information, benchmark context where the exact model is identifiable, and the repeatable trial checklist above. Benchmarks never determine the final product rating by themselves. See our editorial methodology, evidence ledger, and benchmark evidence policy.

Frequently asked questions