mcp-repo-release-tools
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Use this server for any task in this repository that touches release policy: version numbers, the changelog, or branch or commit naming, or [tool.rrt] config. If your client cannot read files directly, that also includes the current .rrt/*.lock.toml state — a file-capable agent gets nothing from the lock-reader tools that Read .rrt/<name>.lock.toml does not already give it. If you were about to run rrt ... in a shell, check here first.
Prefer these tools over shelling out to the CLI for three concrete reasons: (1) most tools return typed JSON (Pydantic models) instead of ANSI-formatted human output you would have to parse — the four lock readers and rrt_config return raw dicts, and rrt_init_run returns captured text, not a model; (2) arguments are passed directly, so a commit subject or branch name never goes through shell quoting — no escaping backticks, quotes, !, or newlines; (3) every mutating tool defaults to dry_run=True, so a preview is what you get unless you explicitly ask for a write. On the CLI the safety is a flag you must remember; here it is the default.
Reach for it when you need to: read or bump a version (rrt_version, rrt_bump); read the changelog (rrt_changelog); check a branch name or a Conventional Commit subject BEFORE you use it (rrt_validate_branch, rrt_validate_commit); inspect resolved config (rrt_config); run a policy check (rrt_doctor, rrt_release_check, rrt_folder_check, rrt_docs_check, rrt_arch_check, rrt_eol, rrt_sync_check); see which dirty paths are generated (rrt_git_dirty_tree) or count commits since a ref without them (rrt_git_commits_since); land a finished branch in one call (rrt_git_land); or run the declared check list against expected exits (rrt_gate_run).
Do NOT use it for: ordinary git (commit, push, rebase, log) — but land a branch with rrt_git_land, never as separate merge, worktree and branch calls; running tests or linters; or rrt subcommands that have no tool here — rrt docs map, rrt docs generate, rrt docs publish, rrt docs inject, rrt tree --check, rrt toc, rrt changelog lint, rrt changelog compare, rrt drift generate, rrt drift check, rrt artifacts --check, rrt arch baseline, rrt arch new, any other --snapshot write, and every rrt-hooks subcommand still require the CLI. Shell out for those; that is the correct surface, not a fallback.
Safety: rrt_bump, rrt_branch_new, rrt_publish_snapshot, rrt_git_land, and rrt_gate_run mutate the repository or run its commands. Call each with its default dry_run=True first, show the user the preview, and pass dry_run=False only after the user confirms. rrt_publish_snapshot force-pushes and additionally requires confirm=True. rrt_init_run also mutates but is the submit target of the rrt_init form, not a tool to call directly — if you are working from a shell, run rrt init instead.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| logging | {} |
| prompts | {
"listChanged": true
} |
| resources | {
"subscribe": false,
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| rrt_configA | Read the repo's resolved [tool.rrt] policy. Version targets, pin targets, changelog file, release branch pattern, folder
rules. Call this before proposing any release-related change so you act on
configured policy rather than assumptions. This returns the resolved config as
structured data — a config-load error comes back as a typed ConfigError rather
than a nonzero exit — but it does NOT run the per-target checks |
| rrt_doctorA | Check whether hooks, workflows, and the artifact-protection lens are wired correctly. Covers pre-commit, lefthook, husky, GitHub Actions workflows, and CI artifact-protection. Use before recommending a hook or workflow change, and when a hook did not fire as expected. Returns one CheckResult per component with ok + severity — no output parsing. |
| rrt_healthA | Return the health check results from .rrt/health.lock.toml. Reflects the last snapshot-writing run, not current state; run the CLI check
to compare against the working tree. If your client can read files directly,
|
| rrt_driftA | Return source drift state from .rrt/drift.lock.toml (file hashes and symbols). Reflects the last |
| rrt_treeA | Return the repository tree snapshot from .rrt/tree.lock.toml. Reflects the last |
| rrt_artifactsA | Return the artifact integrity map from .rrt/artifacts.lock.toml. Reflects the last |
| rrt_versionA | Read the current version of every configured version group from its primary target. Use whenever a version number is about to appear in a commit message, changelog heading, docs string, or release note — this is the authoritative value, not whatever a file happened to say. Read-only. |
| rrt_bumpA | Preview or apply a version bump — use instead of editing version strings by hand. This is the same pipeline as level: major | minor | patch | alpha | beta | rc. dry_run=True by default.
group: restrict the bump to one Runs the SAME pipeline as the |
| rrt_validate_branchA | Check a branch name against this repo's configured type allow-list before you create it. Cheaper than creating it and having the pre-commit hook reject it. Honors [[tool.rrt.branch_types]] and extra_branch_types from config, which a hardcoded regex would miss. |
| rrt_validate_commitA | Check a commit subject against Conventional Commits as this repo enforces it, before you commit. Prefer this over |
| rrt_changelogA | Read the changelog. Use before adding an entry, so you do not duplicate one that is already there,
and before a bump, to see what will be promoted. section='unreleased' (default)
returns parsed pending entries; section='full' returns the raw file. Read-only —
writing an entry needs the CLI or hook workflow ( |
| rrt_branch_newA | Derive a compliant branch name from a type + description. When dry_run=False, checks it out. Use instead of commit_type: a branch type of this repo — the built-ins feat|fix|chore|docs|refactor| test|ci|perf|style|build plus any [[tool.rrt.branch_types]] / extra_branch_types. dry_run=True by default. |
| rrt_git_dirty_treeA | List every uncommitted path, each marked when it matches [tool.rrt.git].generated. Use instead of |
| rrt_git_commits_sinceA | List commits in ..HEAD, leaving out commits that only touch exclude paths. A file that records HEAD can exclude itself, making "one commit behind" its
resting state. Patterns are root-anchored git globs. exclude_generated adds every
[tool.rrt.git].generated pattern. An unknown ref returns |
| rrt_git_landA | Fast-forward onto to branch, remove the branch's worktree, and delete it. One call, one process: three separate git calls cannot do this safely, because a hook can re-dirty the worktree between them. All checks run before any change. Local only — never pushes. Re-running after a failure skips finished steps. restore_generated discards [tool.rrt.git].generated changes in the worktree first. dry_run=True by default. |
| rrt_publish_snapshotA | Force-push a single-commit snapshot of tracked content to a secondary remote. dry_run=True by default. The force-push requires BOTH |
| rrt_eolA | Check the host runtime and project minimum versions against EOL policy. Use before raising or lowering a minimum supported version, and when deciding whether to drop a version from a CI matrix. Read-only — never writes to .rrt/health.lock.toml. Set fetch_live=True to refresh EOL data from endoflife.date instead of the bundled snapshot. |
| rrt_release_checkA | Verify every version target, pin target, and changelog file resolves and agrees, per version group. Use as the pre-release gate and after any edit that touches a version string in docs or config — pin drift is silent otherwise. Read-only — never modifies files. |
| rrt_sync_checkA | List upstream package versions newer than the current project version. Reads the group's [tool.rrt.upstream] package config and queries the configured registry (PyPI/npm/NuGet/crates.io/Packagist). Read-only — never applies a bump. Requires network access to the registry. |
| rrt_folder_checkA | Check the repo layout against [tool.rrt.folders] policy or named built-in templates. Use before creating a new top-level directory or moving a module, so you place it where policy expects. Read-only — never scaffolds files. |
| rrt_docs_checkA | Check whether .rrt/docs.lock.toml is current against source-owned docs. Use after editing any source docstring that feeds generated documentation,
before opening a PR — CI fails on this drift. Read-only — never regenerates or
writes files. If stale, run |
| rrt_arch_checkA | Check the tree against [tool.rrt.architecture] and its findings baseline. Use before adding a module, a unit or a cross-unit import, and before a PR:
the |
| rrt_gate_runB | Run [[tool.rrt.gates]] lines; every line runs and is judged against Each line returns |
| rrt_health_dashboardA | Health overview: Metric summary, health Ring, per-lock status chart, check cards, and detail table. Renders a UI widget for a human to look at — if you need the values to reason over, call rrt_health / rrt_doctor instead. |
| rrt_version_overviewA | Version target map: each configured file, kind, and current version. Renders a UI widget for a human to look at — if you need the primary-target values to reason over, call rrt_version instead. Unlike this dashboard, which reads every configured target, rrt_version returns only each group's primary target; for secondary/pin target consistency use rrt_release_check. |
| rrt_doctor_dashboardA | Doctor check results: pass-rate Ring, per-check Metrics, status cards, and detail table. Renders a UI widget for a human to look at — if you need the values to reason over, call rrt_doctor instead. |
| rrt_tree_dashboardA | Repository tree: snapshot Metric cards, per-directory bar chart, and clean file table. Renders a UI widget for a human to look at — if you need the values to reason over, call rrt_tree instead (it is still a snapshot, not a live check). |
| rrt_initA | Form to initialize rrt configuration — pick target format, preview, then apply. Renders a UI widget for a human to fill in — if you are an agent working from a
shell, run |
| rrt_init_runA | Run rrt init with the given target format. Defaults to dry_run=True for safety. This is the submit target of the rrt_init form — it shells out to
|
| rrt_locks_overviewA | All lock files at a glance: status donut chart, Carousel of lock summaries, and full detail table. Renders a UI widget for a human to look at — if you need the values to reason over, call rrt_health / rrt_tree / rrt_artifacts / rrt_drift instead. |
| generate_prefab_uiA | Execute Prefab Python code in a sandbox and render the result. The code runs in a Pyodide WASM sandbox with full Python support.
Import everything you use. Use the Always use PrefabApp as the outermost context manager — this enables streaming so the UI renders progressively as code is written: For interactive UIs, pass initial state as a dict and use
Available pipes: upper, lower, currency, length, json, round(n), default(val), truncate(n). Charts live in Values passed via Layout patterns:
Args: code: Python code that builds a Prefab component tree. data: Values injected as variables in the sandbox namespace. sandbox: A Sandbox instance. If not provided, a new one is created on each call. |
| search_prefab_componentsA | Search the Prefab component library. Use this tool to look up exact argument names, accepted values, and usage examples before writing component code. The skill covers patterns and layout; this tool has the API details. The query matches component names and descriptions.
Space-separated terms match independently, so
When a query matches a small number of components, full details
(docstrings, args, examples) are shown automatically. For broad
searches, a compact listing is returned instead. Use Args: query: Filter by component name or description. Space-separated terms are OR-matched. detail: Show full docstrings and args. Defaults to automatic (detailed for ≤5 matches, compact otherwise). limit: Max components to return in detail mode (default 8). No limit in compact mode. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| release_workflow | Step-by-step rrt release guide for the given semver bump level. |
| version_strategy | Recommend the correct semver bump level based on a summary of changes. |
| branch_strategy | Suggest a conventional branch type and slug for a task description. |
| commit_message_guide | Draft a Conventional Commits subject line from staged changes and branch context. |
| changelog_entry | Draft a Keep-a-Changelog bullet point from a commit or change summary. |
| config_setup | Step-by-step guide for adding [tool.rrt] configuration to a project. |
| release_readiness | Pre-release verification checklist: health, changelog, version, branch, and CI status. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| resource_version | Current installed version of repo-release-tools. |
| resource_config | Resolved rrt configuration as JSON. |
| resource_config_schema | JSON Schema for [tool.rrt] configuration — discover valid keys and value types. |
| resource_changelog | Full content of the repository CHANGELOG.md. |
| Prefab Generative Renderer | |
| Prefab Renderer (rrt_health_dashboard) | |
| Prefab Renderer (rrt_version_overview) | |
| Prefab Renderer (rrt_doctor_dashboard) | |
| Prefab Renderer (rrt_tree_dashboard) | |
| Prefab Renderer (rrt_init) | |
| Prefab Renderer (rrt_locks_overview) |
TDQS
Scored across 32 tools
Descriptions work hard to separate near-siblings (e.g. rrt_health vs rrt_doctor, snapshot readers vs live checks, dashboards vs data tools, and '_dashboard' widgets that explicitly redirect to data tools). A few families remain easy to confuse — rrt_health/rrt_health_dashboard/rrt_locks_overview/rrt_doctor all touch overlapping lock/health data, and rrt_init vs rrt_init_run differ only by 'run'. But overall an agent can tell what to call.
Nearly everything uses a consistent rrt_ snake_case prefix with clear verb/noun semantics (noun reads like rrt_tree/rrt_config, action verbs like rrt_bump/rrt_validate_commit/rrt_git_land). The Prefab tools (generate_prefab_ui, search_prefab_components) break the rrt_ prefix but are a distinct sub-domain. Minor oddity: rrt_init (form) vs rrt_init_run (submit) naming.
32 tools is heavy and includes substantial redundancy: 4 snapshot-lock readers, 4 UI dashboards, plus dashboard-adjacent read tools. For agent use many dashboard widgets duplicate data already available from the reasoning-oriented readers, so several tools do not clearly earn their place.
Covers the full release lifecycle: config, init, version read, bump, changelog, release/doc/folder/eol/arch/sync gates, branch and commit validation, branch creation, git dirty/commits/land/publish, and gate runs. Gaps are explicitly documented as CLI-only (e.g. tree --check, docs map --check) and are minor, so no real dead ends for the stated purpose.