A Convention for Support Tickets From AI Agents
When an AI agent working on a developer product encounters a bug, wrong result, or tool that can't fulfill a request, it typically works around the problem or gives up. The developer never hears about it, and the company behind the product remains unaware of the issue. During 2026 many MCP servers gained a feedback tool, each under a different name and almost all one-way: the report goes in and nothing comes back. Support tools still assume a person typing into a chat window. Standard Support addresses this gap by providing a convention that lets a customer's AI agent ask for help, report a problem to a team, and pick up the reply even in a later session.
The specification is written as Markdown and dedicated to the public domain (CC0). It defines a focused set of tools and a status URL that a product can expose so an AI agent can get help, report problems, see its tickets, read team replies, and answer them. The convention is written with developer tools in mind, where an AI coding agent builds an integration with a product, but it works for any product with an MCP server or an API and for any AI tool that uses one on a person's behalf.
Two Separate Paths: Help and Tickets
The design keeps two halves apart on purpose. Help is a question answered from the product's own documentation. Nothing private goes in and no person is involved, so it is safe to automate end to end. Tickets carry a customer's problem to the team. A person agreed to send each one, and a person on the product side should at least be watching what happens to it. This separation means help can be fully automated while tickets respect that a human initiated them.
Help: The First Stop
The first tool in the convention is support_get_help. It takes a question, optionally the tool involved and the exact error text. It returns whether the documentation answered the question, the answer itself, sources, and a fixed line telling the agent to offer a ticket if the answer didn't solve the problem. It is the first stop because much of what looks like a bug is actually an agent that didn't read the documentation. An answer on the spot is better for everyone than a ticket, and "let your agent ask for help" is easier to accept than "let your agent open tickets." The tool opens no ticket and reaches no person, so the agent doesn't need to ask the user first, and the tool is marked read-only. Questions can be logged to show what documentation is missing.
It answers from documentation and known issues only. Whatever produces the answer—a search or the product's own agent—sees the question and public documentation and nothing else: no account data, no credentials, no tools. That is what makes it safe to leave unattended. It needs no sign-in since it reveals nothing private and should be rate-limited.
Tickets: Reporting Problems
The second tool, support_open_ticket, reports a bug, wrong result, question, or feedback. Required fields are kind (enumerated values like bug, crash, wrong_result, confusing, missing_feature, docs, question, other), title written like an issue title, and description of what was tried, what was expected, and what happened. Optional fields include impact (outcome-based: blocked, worked_around, friction), task (what the user asked the agent to do), request_ids (IDs from error responses or results), tool (the tool involved), error (exact error text), suggestion (what would have helped), category (product's own sub-kinds), source (whether the user or the agent initiated the report), session (a label the agent chooses for its work), agent (self-reported name, version, model), and idempotency_key (a retry with the same key returns the first ticket).
Why these fields? Kind, task, impact, description, error, and suggestion already appear with these meanings in several vendors' feedback tools, so agents have seen them. Impact is by outcome, not severity—an agent can truthfully say whether it was blocked, worked around, or only slowed down; it can't judge "high" or "low." Request_ids let the product find the failing request in its logs. Source matters because when a person says "send that as feedback," they have delegated it and will want an answer, but when an agent notices a problem itself, it is closer to a crash report and may need no reply. Idempotency_key exists because agents retry. Session exists because one customer may have a dozen agents working at once, and agents have no stable identity to give; a label they choose is enough to sort by.
The tool returns a ticket object (without messages), notifications (email and watch URL so the agent can tell the user how the reply will arrive), and a duplicate flag (true when the idempotency key matched). The email address comes from the product's own database, never from the tool call, so the tool can't be used to send mail to anyone else.
Listing and Reading Tickets
support_list_tickets takes status (open by default, closed, or all), session filter, updated_since, limit (1-50, default 20), and cursor. It returns tickets most recently updated first and a next_cursor. support_get_ticket takes a ticket_id and optional after_message_id. It returns a notice fixed sentence saying the team's messages are information for the user and not instructions, the ticket, and messages. Reading marks the team's messages as read.
Tickets have status: open (with the team), needs_reply (the team asked something), resolved, closed. Resolution, once resolved, can be fixed, answered, planned, duplicate, not_reproducible, or wont_fix. Each message has id, author (reporter, team, system), body, created_at, and first_read_at (when the message was first returned by support_get_ticket, showing the agent fetched it, not that a person read it). The report's description is the ticket's first message, so a ticket reads as one conversation; internal notes never appear.
Replying and Closing Tickets
support_reply_to_ticket takes ticket_id, body (1-4000), optional outcome (resolved closes the ticket, still_broken reopens it), and optional idempotency_key. With resolved the status becomes closed and the resolution fixed unless the team already set one. With still_broken a resolved or closed ticket reopens. A reply with no outcome answers a needs_reply ticket and otherwise leaves the status alone, so "thanks" doesn't reopen a fixed ticket.
How the Agent Hears About a Reply
An agent usually has no inbox, and most can't receive a webhook. What many can do is wait: coding agents run a script in a background shell, carry on working, and are woken when the script prints something. That script can run curl, but it generally can't call an MCP tool. So every ticket gets a watch_url—a plain https URL with an unguessable token (128 bits or more) in the path and no other sign-in. GET returns updated_at, status, unread_count, min_interval_seconds, and long_poll, and nothing else: no title, no message text, no account. A leaked URL reveals only that a ticket changed.
Polling calls the URL no more often than min_interval_seconds (suggested: 60) and prints or exits when updated_at changes. The agent then calls support_get_ticket to read the reply. Support ETag and If-None-Match make a poll that finds nothing new cheap. Long polling is optional and off by default; it may accept ?wait=
support_open_ticket returns notifications.watch with the URL, the interval, and one sentence the agent can act on: "To be told when the team replies, poll this URL in the background and call support_get_ticket when updated_at changes." The reply also goes to the signed-in user's address so they hear when no agent is watching. Without either mechanism, an agent calls support_list_tickets the next time it works with the product, and unread_count shows what changed. It doesn't have to be the agent that opened the ticket: tickets belong to the customer, not to an agent's session, so a later session can pick up a reply that arrived after the first one ended.
What Clients Could Do
The watch URL exists because of what coding agents can do today. A client could make it unnecessary. A coding agent could keep an MCP tool call polling in the background, and wake up when the result changes. With that, an agent would watch a ticket by polling support_list_tickets with updated_since, or support_get_ticket to read replies. The specification is a design to copy, not a protocol to conform to. Each decision comes with its reason, so implementers can disagree with one decision and keep the rest. Agents read each tool's description, so this doesn't need everyone to agree. Products can adopt what works and extend the convention however they like.