docs: prefer the Binjovi MCP server over binjovictl #713

Merged

binjovi-bot merged 1 commit from docs/prefer-binjovi-mcp into trunk

2026-09-13 16:50:55 +00:00

sean commented

2026-09-13 16:35:52 +00:00

Owner

The contributor guidance named binjovictl as the release train, so a session reaches for a CLI and parses its text. The Binjovi MCP server is the same authority through a typed interface, and it is strictly better for this:

It returns structured evidence rather than lines to grep.
request_release and get_operation_status each wait for up to 25 seconds on their own, so wrapping a CLI in a for loop with sleep 30 is a worse version of a call that already waits. I did exactly that several times today before noticing.
list_release_candidates and list_deployment_targets give the base revision and the targets that request_release requires, so they do not have to be guessed.

AGENTS.md (which CLAUDE.md symlinks to) now names the MCP calls first, says where request_release gets its arguments, keeps the candidate_catalogs_diverged explanation attached to releasing one candidate by build id, and keeps binjovictl as the fallback when the server is unreachable, noting that it derives the base revision and targets for you.

The one gap is stated rather than hidden: no MCP call waits on a Build, so waiting for a Build to succeed before releasing it is still a poll.

tests/current-ci-guidance.sh gains an assertion that the guidance keeps naming the server first. Verified red against the previous text and green after.

Scope is AGENTS.md only. docs/DEVELOPMENT-WORKFLOW.md still documents the binjovictl line that the same check pins, and changing that is a separate edit.

https://claude.ai/code/session_01XdBRc9CHgFhAxCyHbdVtt6

The contributor guidance named `binjovictl` as the release train, so a session reaches for a CLI and parses its text. The Binjovi MCP server is the same authority through a typed interface, and it is strictly better for this: - It returns structured evidence rather than lines to grep. - `request_release` and `get_operation_status` each wait for up to 25 seconds on their own, so wrapping a CLI in a `for` loop with `sleep 30` is a worse version of a call that already waits. I did exactly that several times today before noticing. - `list_release_candidates` and `list_deployment_targets` give the base revision and the targets that `request_release` requires, so they do not have to be guessed. `AGENTS.md` (which `CLAUDE.md` symlinks to) now names the MCP calls first, says where `request_release` gets its arguments, keeps the `candidate_catalogs_diverged` explanation attached to releasing one candidate by build id, and keeps `binjovictl` as the fallback when the server is unreachable, noting that it derives the base revision and targets for you. The one gap is stated rather than hidden: no MCP call waits on a *Build*, so waiting for a Build to succeed before releasing it is still a poll. `tests/current-ci-guidance.sh` gains an assertion that the guidance keeps naming the server first. Verified red against the previous text and green after. Scope is `AGENTS.md` only. `docs/DEVELOPMENT-WORKFLOW.md` still documents the `binjovictl` line that the same check pins, and changing that is a separate edit. https://claude.ai/code/session_01XdBRc9CHgFhAxCyHbdVtt6

sean added 1 commit

2026-09-13 16:35:52 +00:00

docs: prefer the Binjovi MCP server over binjovictl

binjovi/ci Binjovi completed the frozen plan

Details

1aa8799a48

The contributor guidance named binjovictl as the release train, so every
session reached for a CLI and parsed its text. The MCP server is the same
authority through a typed interface: it returns structured evidence, and
its release and deploy calls wait for up to 25 seconds of their own, so a
polling loop around a CLI is a worse version of a call that already waits.

The guidance now names the MCP calls first, says where request_release
gets its base revision and targets, and keeps binjovictl as the fallback
when the server is unreachable. A check keeps the preference from
regressing.

The one gap is stated rather than hidden: no MCP call waits on a Build, so
waiting for a Build to succeed is still a poll.

Claude-Session: https://claude.ai/code/session_01XdBRc9CHgFhAxCyHbdVtt6

binjovi-bot merged commit 1aa8799a48 into trunk

2026-09-13 16:50:55 +00:00

binjovi-bot deleted branch docs/prefer-binjovi-mcp

2026-09-13 16:50:56 +00:00

No reviewers

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

sean/seanfarm!713

No description provided.

Rows
Columns