Inbox Monster publishes a Model Context Protocol (MCP) server, so an AI assistant can read and act on your account directly rather than you exporting a report and pasting it into a chat.
If you use the Claude apps, you don't need this article. The connector is listed in Claude and takes two clicks to set up. See Connecting Inbox Monster to Claude.
This article is for teams running an agent of their own: a command line client, an agent framework, or something built in-house, often alongside the ESP you already send from.
How access works
There is no API key to generate or paste. The server uses OAuth, so your agent sends the user to Inbox Monster, the user logs in and picks a team, and a token is issued for that team.
Three things follow from that, and they are worth knowing before you build:
Access is scoped to one team. If your account covers several brands, the token reaches the team chosen at approval and no other. To work with a different team, disconnect and reconnect.
The agent inherits the user's permissions. It cannot reach anything the team cannot reach, and it cannot do anything the approving user could not do in the platform.
The connection needs a person. Someone logs in and approves it, and the agent then acts as that user on that team. There is no service account, so an agent that has to run with no person behind it is not something the server supports today.
Before you start
An Inbox Monster login with access to the team you want the agent to work with.
An MCP client that supports remote servers over OAuth. Most current clients do, including Claude Code, Codex, Cursor and the OpenAI Agents SDK.
Connection details
Server URL |
|
Transport | Streamable HTTP |
Authentication | OAuth, on first use. Your client registers itself, so there is no client ID to request from Inbox Monster and no key to paste. |
Connecting
Claude Code
claude mcp add --transport http --scope user inboxmonster https://mcp.inboxmonster.com/mcp
Run /mcp, select inboxmonster and authenticate.
Codex
codex mcp add inboxmonster --url https://mcp.inboxmonster.com/mcp codex mcp login inboxmonster
Other clients
Add https://mcp.inboxmonster.com/mcp as a remote MCP server over streamable HTTP, then sign in when prompted. Your client has to support MCP OAuth.
Configuration file keys differ by client. Cursor uses mcpServers with url; VS Code uses servers. Check your client's documentation for the exact shape.
Approving the connection
The client opens an Inbox Monster page. Log in if you are not already signed in.
Choose the team the agent should work with, then click Allow.
Ask the agent to list your recent renderings. Real campaign names mean the connection is live.
The access level comes from the client, not the consent screen Read and write access is part of the request your client makes. The team is the user’s choice; the access level is your client’s. If you want an agent that reads and never writes, restrict it in your own client rather than expecting the consent screen to do it. |
What the server covers
The tools fall into three areas. Your agent sees only the tools the team and the approving user already have access to, so your list may be shorter than this.
Creative Rendering
Task | What it does |
Submit a creative | Creates a rendering from HTML and generates previews. Adds new versions to an existing rendering. |
Browse | Lists renderings and retrieves a specific version. |
Previews | Returns device previews, lists available devices and re-runs an individual device. |
Checks | Returns link validation, image, spelling, accessibility and content analysis results. |
Custom QA | Runs your rule library against a rendering and returns the results and aggregate pass and fail counts. |
AI summaries | Generates and returns the AI summary for a campaign. |
Sharing | Returns a share URL for a rendering and updates its share settings, so a reviewer without a login can see it. |
Campaigns and engagement | Lists campaigns, returns campaign detail and returns engagement data including browser and country breakdowns. |
Deliverability
Inbox placement, blocklists, spamtraps, reputation, DMARC and Monster Score. Agents can create delivery tests and read seedlist results, spamtrap hits by domain, reputation data and active blocklistings.
Administration
Managing domains, IPs and users on the team, including adding a domain or IP and changing an IP label.
Your ESP
If you have connected an ESP to Inbox Monster, the agent can list the connected integrations and browse the templates, sending domains, IP pools and message types inside them. That is what lets an agent fetch a template from your ESP and submit it for rendering in one step, rather than someone exporting HTML by hand.
Read-only and interactive tools
Tools divide into two kinds, and the distinction matters more for an agent you have built than for a person chatting:
Read-only tools fetch information and change nothing: campaign results, rendering previews, engagement figures, blocklist and spamtrap hits.
Interactive tools act on the account, such as creating a rendering, running a Custom QA check or adding a domain.
Hold the interactive ones for approval unless you have a reason not to. An agent that loops will call them more often than you expect.
What counts toward your usage
Renders generated by an agent are billed the same as renders generated in the platform A device preview is a device preview whether a person clicked the button or an agent called the function. Re-running a single device counts too. |
This matters more with an agent because an agent can loop. An instruction like "keep adjusting the HTML until every device renders cleanly" can generate a large number of previews quickly. Two things help:
Give the agent a device list rather than letting it request the full library. Most teams review 20-30 devices per email.
Check your usage after the first few agent runs so you know what a typical run costs you.
Example prompts
"List the renderings from the last seven days and tell me which ones have failing links."
"Run Custom QA on the latest version of the holiday campaign and summarise the failures by rule."
"Pull the previews for this rendering on iPhone and Gmail, and tell me where the layout differs."
"Fetch the welcome email template from our ESP, submit it for rendering, then report any image over 102 KB."
"Compare engagement on the last three campaigns by country."
Disconnecting and switching teams
Remove the server from your client, or revoke the token. The agent stops reaching your account straight away. To work with a different team, disconnect and reconnect, then choose the other team at approval.
Troubleshooting
What you see | What to check |
The agent has no Inbox Monster tools | The client has not loaded the server. Confirm the URL and restart the client. Most clients read their configuration only at startup. |
The approval page does not open | The client is not handling the OAuth redirect. Check it supports remote MCP servers rather than local ones only. |
Authentication errors after it was working | The token was revoked or the connection replaced. Reconnect and approve again. |
Connects but returns no data | The token belongs to a team with no data, or the approving user cannot see it. Check which team was chosen. |
Some tools are missing | The team or the approving user does not have access to that product area. |
Previews time out | Rendering across a device library takes time. Ask for previews in a second call rather than expecting them in the response to the submission. |
Frequently asked questions
Do I need an API key?
No. You log in, choose a team and approve. A token is created for you.
Can an agent connect without a person approving it?
No. Someone logs in and chooses a team, and the agent acts as that user. There is no service account, so a pipeline that runs with nobody present needs a person to make the connection first.
Can one agent reach several teams?
No. A connection covers a single team. Reconnect to switch.
Can the agent do more than the person who approved it?
No. It inherits that user's permissions on that team.
Is this the same as the AI features inside Inbox Monster?
No. AI analysis, AI summaries and Custom QA run inside the platform. The server lets your own agent reach your Inbox Monster data.
