Skip to content

Deploy tokens

A deploy token is what a coding agent presents to seylo-mcp. It is a capability, not a login. It names one cluster and one namespace, carries a list of capabilities, and expires. It can’t approve a request, read a Secret value, reach another namespace or widen its own scope.

Only an admin can issue, list, renew or revoke tokens. All of it happens on Admin → Clusters, on the Coding agents card.

  • One namespace on one cluster. Each token is bound to exactly one namespace, and that namespace must be open to coding agents when the token is issued. For a second namespace, issue a second token.
  • The scope is checked on every call, against the cluster’s policy as it is now. Seylo doesn’t trust what was true when the token was issued.

The admin ticks capabilities one at a time. If none is ticked, the token is read-only.

Capability Tools Default Notes
list_targets list_targets Yes
describe_target describe_target, diagnose Yes
operation_status operation_status Yes
logs logs Yes
observe The 22 observability read tools Yes Not on tokens issued before 2026-10-04
validate validate No Changes nothing
list_config list_config No Secret names and key names only
request_secret request_secret No Creates work for a person
dashboard_propose dashboard_propose No Creates drafts for a person to decide on
scrape_target_propose scrape_target_propose No Files requests into the approval queue
deploy deploy No Changes the cluster. Each applied deploy counts against the daily cap
rollback rollback No Changes the cluster. Each applied rollback counts against the daily cap
put_config put_config No Changes the cluster
propose_deploy propose_deploy No Files plain manifests as one deploy for a person to approve. Never a secret value, never cluster-scoped. Counts against the daily cap. See deploy from the coding agent

The full list is on the tool reference.

Default Range
Expiry 30 days 1 to 90 days
Daily deploy cap 20 0 to 1000
Label 1 to 80 characters
Read calls 60 per token per minute Fixed
Write calls 10 per token per minute Fixed
Dashboard drafts waiting 10 per token Fixed

The daily deploy cap limits deploy and rollback per UTC day, and resets at 00:00 UTC. The cap counts deploys that were applied: each applied deploy or rollback counts once, and a refused or failed attempt counts zero. Reads don’t count against it. A cap of 0 issues a token that can carry deploy or rollback but can’t use either.

Every tool call is rate-limited, per token, in two separate buckets:

  • Reads, 60 calls a minute. Every tool that changes nothing: the 22 observability tools, logs, diagnose, validate, describe_target, list_targets, list_config and operation_status.
  • Writes, 10 calls a minute. deploy, rollback, put_config, request_secret, dashboard_propose and scrape_target_propose.

Each bucket is a rolling one-minute window, and the two are counted independently, so a busy metrics loop can’t use up the room a deploy needs. A call that the tool itself refuses, such as a deploy blocked by policy, still counts. The daily deploy cap applies on top of the write limit: the cap bounds how much a token can change in a day, the write limit how fast it can try. A call over either limit is refused with a message that names the family, the limit and how many seconds to wait. See troubleshooting. list_targets and observability_status both state the two limits, so an agent can pace itself.

A token stops working when it expires, and its row shows expired. A caller holding it gets the same refusal as for any token that isn’t valid. An admin can renew a token that is still active, or that expired within the last 30 days.

Follow issue a token. The plaintext token appears once, on the screen after you press Issue token. Seylo stores only a hash of it. No screen, API or administrator can show it again.

Tokens start with seylo-deploy_. If your secret scanner lets you add patterns, add that prefix.

Issuing a token writes an audit row that records who issued it, the target, the capabilities, the expiry and the cap. The token itself is never in it.

The Coding agents card lists every token in your organisation:

Column Shows
Label The label, and issued by with the issuer’s email
Target cluster / namespace
Capabilities One badge per capability. Tokens that can change the cluster also show deploys used today against the cap, such as 3/20 deploys today
Expires expires in N days (expires within a day on the last day). At 7 days or fewer it also shows an expires soon badge. Once expired, expired N days ago (expired today on the day). Hover for the exact time
Last used The last call of any kind, including a refused one, or never
Status active, expired or revoked. A token that was replaced by renewing it shows renewed

The admin who issued a token, or who last renewed it, gets one email when 7 days or fewer are left, with the subject Deploy token "<label>" expires in N days. It is sent once per token, never for a revoked token, and not at all for a token issued for 7 days or fewer. The coding agent holding the token is told nothing new: after expiry it gets the same refusal as for any token that isn’t valid.

An operation started by a coding agent is recorded against the person who issued the token, with the token’s label in the summary. When someone asks who deployed something, the answer is a person. The agent is how they did it.

The issuer also receives an email for each deploy and rollback, and for each deploy refused by policy, by the dry run or because another controller owns its target. A burst of refusals usually means an agent that is confused, or one that is following instructions it shouldn’t. The issuer is the person best placed to tell which.

Renewing gives a token a fresh lifetime without choosing its capabilities again. Renew appears on a token’s row while it is active, and for 30 days after it expires.

  1. On the Coding agents card, press Renew on the token’s row.
  2. Read the confirmation, Renew this deploy token?, and press Renew token.
  3. Copy the new token now. It appears once, on the same screen as an issued token, with ready-made connection commands. Seylo can’t show it again.
  4. Replace the old token in the coding agent with the new one, for example by running one of the commands on that screen.

What renewing does:

  • It issues a new secret for the same cluster, namespace, capabilities, label and daily cap.
  • The new token lasts as many days as the old one was first issued for, at most 90, counted from now.
  • The old token is revoked at the same moment. There is no overlap, so any coding agent still using it is refused until you give it the new one. Its row then shows renewed.
  • The admin who renews becomes the token’s issuer. That admin gets the next expiry reminder and the email for each deploy and rollback.
  • The daily cap doesn’t reset. Deploys and rollbacks already applied today still count against it after renewal.
  • The capabilities stay exactly as they were. Renewing doesn’t add observe to a token issued before 2026-10-04. Rotate for that.

Renew is refused in these cases, and the dialog shows why:

Message begins What to do
This token was revoked, and a revoked token cannot be renewed Issue a new token for the same namespace
This token expired more than 30 days ago, so it can no longer be renewed. Issue a new token, and choose its capabilities again
This token was revoked or renewed a moment ago, so it was not renewed again. Someone else renewed it first. Refresh the list to see its replacement

Renewing a token also needs its namespace to still be open to coding agents, and its cluster to still be in Seylo.

Press Revoke on the token’s row and confirm Revoke token.

  • The token stops working immediately. Its next call is refused.
  • Revoking can’t be undone.
  • Nothing already deployed is touched.
  • The token’s history is kept. Revoking stops future use without erasing what the token did.

Revoke a token when it is lost or leaked, when the person or machine it was issued to no longer needs it, or when you want it never to work again.

To keep the same capabilities, renew the token instead. To change its capabilities, replace it:

  1. Issue a new token for the same namespace, with the capabilities you want now.
  2. Update the client with the new token. In Claude Code, add the server again with the new token.
  3. Check the connection: list_targets should list the new capabilities.
  4. Revoke the old token.

Rotate before a token expires to avoid a gap. Rotation is also how an older token gets observe or any other capability. A token’s capabilities can’t be edited, and renewing doesn’t change them.

If an admin closes the namespace, every token for it is refused from its next call. Nobody has to revoke anything.

Closing a namespace is not revoking its tokens. If the namespace is opened again, its tokens that haven’t expired or been revoked work again. If that isn’t what you want, revoke them as well.

A token also stops working if its cluster is removed from Seylo.