Postmark
Transactional email from your own Postmark account, through Postmark's MCP server, which you host.
Setting this up needs a developer or your IT team. Postmark offers no server that Zero Human OS can connect to directly, so someone at your company runs Postmark's own server, with a bridge in front of it, on a machine your company controls. Hand them the Setup section of this page, your Postmark server API token, and the address mail should come from. They give you back two things: the bridge's web address, and the bearer token it expects. With those, you connect it in the portal yourself, as Connect it shows.
What a member can do
With Postmark connected, a member can send email from your Postmark server, singly, in batches or from your templates; manage templates; search sent messages and bounces, and diagnose a delivery; and read your delivery statistics. Which of these it has is what the server you host offers and what the task names.
Postmark's own server, @activecampaign/postmark-mcp, serves the tools below, under these names. A bridge in front
of it may offer fewer of them, never more.
What it reads
| Tool | What it reads |
|---|---|
listTemplates |
The templates on your Postmark server: name, id, alias, subject and type |
getTemplate |
One template's subject, HTML and text bodies, and layout |
validateTemplate |
Checks a template's syntax, and renders it against test data, without saving anything |
searchOutboundMessages |
Sent messages, by recipient, sender, subject, tag, status, message stream or date |
getMessageDetails |
One sent message's delivery details and events: delivered, opened, clicked, bounced |
diagnoseDelivery |
Why a message may not have reached a recipient: its messages, suppressions and bounces together |
searchBounces |
The bounce log |
getBounceDump |
The raw exchange with the receiving mail server for one bounce |
listSuppressions |
The addresses a message stream does not send to |
getDeliveryStats |
Open, click, bounce and spam rates, and their breakdowns |
getServerInfo |
Your Postmark server's settings |
listWebhooks |
The webhooks on your Postmark server |
What it writes
| Tool | What it changes |
|---|---|
sendEmail |
Sends one email, to up to 50 recipients |
sendEmailWithTemplate |
Sends one email from one of your templates |
sendBatch |
Sends up to 500 separate emails in one call |
sendBatchWithTemplate |
Sends one template to up to 500 recipients, each with its own values |
createTemplate |
Adds a template |
editTemplate |
Changes a template in place, which cannot be undone |
deleteTemplate |
Deletes a template, which cannot be undone |
activateBounce |
Lets an address that bounced receive mail again |
createSuppressions |
Stops a message stream sending to up to 50 addresses |
deleteSuppressions |
Lets up to 50 suppressed addresses receive mail again |
createWebhook |
Adds a webhook: Postmark calls its address on the events you choose, until it is deleted |
deleteWebhook |
Deletes a webhook |
Name each Postmark tool a task needs, such as postmark.sendEmailWithTemplate and
postmark.searchOutboundMessages, rather than all of Postmark: a tool the task does not name is refused, and a
task that can only send from a template cannot write an email of its own.
This is for mail your company sends. A member's own Zero Human address, and the Inbox on its member page, only receive.
Who connects it
Postmark offers no hosted MCP server. Its own server, @activecampaign/postmark-mcp, runs next to whatever uses
it, so to connect Postmark you run that server yourself, behind an HTTPS address that Zero Human OS calls with a
token of your choosing. Your Postmark server token stays on your server; Zero Human OS only ever holds your bearer
token.
The sign-in is that token: Zero Human OS sends it as Authorization: Bearer <your token> on every call to the bridge
you host, and the bridge refuses any call without it.
Mail goes out from your Postmark server, from a sender your Postmark account has confirmed: the one a call names, or the default sender you set on your server.
Connect Postmark on the task that sends, not on a member, team, role or the enterprise. A connection on any of those also reaches the chat of every member it covers, where a send is a single approval away from going out without the task's skill, inputs or checks. A task's own connection never reaches chat.
What waits for you
Nothing sent through Postmark waits for you unless the task declares a gate on the tool that sends. To see each email before it goes, gate that tool: the run waits with the exact message it would send, and nothing is sent until you approve it on Gates.
A gate is part of the task. In the portal, open the task and edit it. Under Gates, fill in Tool with the
tool that sends, such as postmark.sendEmailWithTemplate, Owner with ceo and Approval with specific, so
that every call asks, then choose Add gate and save. Saving publishes a new version of the task. If you write
tasks through the Tasks API instead, the same gate is this entry in the version's gates:
{ "tool": "postmark.sendEmailWithTemplate", "owner": "ceo", "approval": "specific" }
The same holds for the tools that change templates, suppressions and webhooks: nothing they change waits for you unless the task gates them. Gate them, or leave them off the task.
Setup
Run Postmark's MCP server
This part is for the developer or IT person who runs it. It is the same work as running any small web service.
- Run
@activecampaign/postmark-mcp, the package published by Postmark's owner, under exactly that name. In 2025 a copy published under the bare namepostmark-mcpblind-copied every email it sent to its author, and npm has since removed it; never install a package without the@activecampaign/scope. It needs your Postmark server API token, a default sender your Postmark account has confirmed, and a message stream (outboundfor transactional mail). - Postmark's server only talks to a program on the same machine (MCP over standard input and output), so Zero
Human OS cannot reach it as it is. Put a bridge in front of it: a small web service that passes each request on
to Postmark's server and serves it over MCP's Streamable HTTP transport, the way MCP servers are reached over
the web. Any MCP proxy that turns a standard-input server into a Streamable HTTP one will do, as long as it serves
a public HTTPS address and refuses any request whose
Authorizationheader is notBearer <your token>. Choose a long random token. - If your bridge can, offer only the tools your tasks need. What the bridge does not offer, no task can call.
- Hand back the bridge's HTTPS address and the token. Whoever connects it in the portal needs both.
Connect it
- In the portal, on Tools, choose Add tool. Layer: Task, and Bind on: the task that sends.
- Kind: MCP. Tool type:
postmark. - MCP server URL: your bridge's address.
- Auth: Bearer token / PAT, and your bearer token.
- On the task, name the Postmark tools it needs, and declare a gate on each one that sends.
- Check it. Run the task with your own address as the recipient. Approve the gate, and the run page shows the call and Postmark's reply; the message appears in Postmark's Activity.
To disconnect Postmark, delete the binding on Tools in the portal. A run of a task that names a Postmark tool then stops until one is connected again. To cut the bridge off as well, stop it or change the token it expects; to stop all mail from that Postmark server, revoke its server API token in Postmark.
Limits and cost
What one call can do is Postmark's own limit: one email goes to at most 50 recipients, a batch holds at most 500 messages, a search returns at most 500 messages, and a call adds or removes at most 50 suppressed addresses. Postmark keeps bounce records for 45 days, and a bounce's raw exchange for 30. Your Postmark account's own sending limits apply, and so does anything stricter that your bridge sets.
Postmark's sending is billed to your Postmark account, at your Postmark plan's prices. The bridge has no Zero Human charge of its own: its calls happen inside runs, and your plan counts those runs as it counts any other. Running the bridge costs whatever the place you host it charges.
Troubleshooting
All of these show on the run page: a run that stops says why at the top, and Show log opens its full record. The words in code are what the page shows, so you can search this page for them.
| What you see | Why | What to do |
|---|---|---|
The run stops and says no Postmark connection reaches it (tool_not_bound:postmark) |
No Postmark connection reaches this run | Bind it on this task, or at a layer the run reaches |
The run carries on without a Postmark tool, and its log says the tool was not offered (tool.not_offered) |
Your bridge did not answer when the run started, or offers no tool by that name | Ask whoever runs your bridge to check it answers at its public address, and that it offers the tool by that name |
| Every Postmark call fails with a sign-in (auth) error | The bearer token in the binding is not the one your bridge expects | Edit the binding and paste the token again |
| Postmark refuses the sender | The sender is not confirmed in your Postmark account | Confirm the sender signature or domain in Postmark |
| A send goes out without an approval | The task declares no gate on that tool | Add a gate on it to the task |