Repository navigation
Releases
v0.8.1 — packaging hotfix: build-synced agent docs
v0.8.1 fixes a packaging defect in v0.8.0: the wheels published to PyPI and attached to the v0.8.0 release carried agent documentation from the 0.6.x era, so specialists installed from a wheel — or from the read-only package data of a system install — received outdated instructions. The 0.8.0 code itself was correct; only the packaged documentation copies were stale. This patch release rebuilds the package with the documentation synced at build time and adds a guard test so the drift cannot recur silently.
What changed
- Agent docs shipped in the wheel are now build-synced: the build copies the source-of-truth agent files (
JTM_AGENT.md,jira-tempo-reports.skill.md,jtm-jira-tempo-reports.agent.md) into the package before assembling, so what you install matches the repository. - A guard test fails the build if the packaged copies ever diverge from the source files again.
- The packaged
JTM_AGENT.mdin 0.8.1 includes the full 0.8.0 tool inventory, including thereport_gapsritual. - No Python code changes: 0.8.1 is functionally identical to 0.8.0; only packaging and version metadata changed.
Verification
Built from commit 9ef3557 on main. The wheel's integration/JTM_AGENT.md was read back from the built artifact and contains 3 occurrences of report_gaps. SHA256:
jira_tempo_mcp-0.8.1-py3-none-any.whl—8eb6904460d239aa003c3083685c8e67df53d67eaa893f249e11f193cf18a607jira_tempo_mcp-0.8.1.tar.gz—b0bb2313f6d60a49d813cc3a98e3195f1f2b31cdcef4b8c95875b2681f75f1ec
Compatibility and rollback
Drop-in replacement for 0.8.0: same tool inventory, same configuration, no migration steps. v0.8.0 stays published (not yanked) — 0.8.1 supersedes it; pin 0.8.0 only if you must stay frozen.
Both the wheel and the sdist are attached below.
jira-tempo-mcp v0.8.0
jira-tempo-mcp 0.8.0 adds three tracking-workflow tools and one repository guard fix. You can now log a whole day of work in a single call, check which weeks of the year are missing time entries or reports, and get per-issue or per-day totals out of list_worklogs without any post-processing.
Highlights
- Batch worklog creation (
create_worklogs, P7). One call takes up to 20 entries (issue_key,time_spent, optional comment and start date). Every entry is validated before the first write: an invalid batch commits nothing and makes no network call at all. Valid entries are written sequentially with no retries (the Tempo POST is not idempotent, so a retry could double-log time); failures are reported per entry and never block the rest, and the response closes with a written/failed summary. - Reporting gap detector (
report_gaps, P8). A strictly read-only scan that lists weeks since January 1st (look-back capped at 26 weeks, older values clamped and disclosed) where tracked hours > 0 but the weekly report file is missing. On its first run against the owner's 2026 data it found 3 real gaps. Per-week coverage comes back with tracked hours and a summary line. - Server-side aggregates (
list_worklogs, P5). New optionalgroup_by(issueorday) andinclude_totalsparameters aggregate exactly the rows the tool already renders. The default output is byte-identical to 0.7.0 (guarded by tests), so existing callers see no change. - Virtualenv guard fix.
.gitignorepatterns ending in a slash silently miss a.venvsymlink (the abc55c8 incident class); the patterns are now slash-free and a CI test fails if a virtualenv path is ever tracked again.
Verification
make ci green: 616 tests passed (568 in 0.7.0), verified on the release commit before tagging.
Compatibility and rollback
No breaking changes: all new parameters are optional and defaulted, the MCP tool list grows by two entries (create_worklogs, report_gaps). Docker users stay on the ghcr.io/korrnals/jira-tempo-mcp:0.8.0 pin; rolling back is a pin change to 0.7.0 plus docker compose pull. Wheel and sdist are attached to this release.
v0.7.0 — observability: week_status + health_check, live preview, strict validation
This release makes the weekly-report workflow observable and self-checking: two new read-only tools (week_status, health_check) let you and the agent see what is actually tracked and whether the server is configured correctly — before a report is ever generated.
Highlights
- Strict report-tool validation. Generator tools (
generate_weekly_report,generate_team_report,generate_tasks_report,preview_report_template) now reject unknown parameters with an explicit error. Previously a misconfigured call (e.g.date_fromon the weekly generator) was silently ignored — the report could be generated for a different week than intended. That entire failure class is closed: verified live, a wrong parameter now stops the call instead of producing a wrong file. week_statustool (read-only). Per-day tracked hours over the Mon–Sun data window of the target week, totals vs theexpected_hoursnorm, per-day overtime flags. Tracked time only — issue keys and comments are never included, so it is safe to call at any moment.health_checktool. Server/Python/MCP SDK versions, configuration paths, and credential presence flags (JIRA_PAT: set) — values are never returned. Optionalprobe_apis=truedoes a read-only Jira/Tempo probe and reportsok/FAILEDper API.- Live template preview.
preview_report_templategainedlive=true+target_date: renders the real worklogs of the target week into the tool response without writing any file — a dry run of the exact fetch path the generators use. Mock mode remains the default. include_droppedtransparency flag. The txt renderer drops title-echo lines; withinclude_dropped=truethe dropped lines are now listed in the tool response (the report file stays byte-identical), so "where did my line go" is answerable without guessing.- Canonical output contract pinned. The
defaulttemplate description now states the canonical report format, backed by a golden byte-identity gate in CI — any unintentional drift in the builtin template's render fails the build.
Verification
Full local gate green on the release commit: ruff lint clean, mypy clean (23 source files), 568 tests passed, sdist/wheel built; the live V-series checks (wrong-week rejection, live preview, week_status, strict error) were exercised against the real Jira/Tempo instance. Docker image pushed to ghcr.io/korrnals/jira-tempo-mcp:0.7.0 (+ latest), PyPI and npm wrapper published at 0.7.0.
Compatibility
No breaking changes: all new capabilities are additive tools/flags; existing tool calls keep their semantics (unknown keys now error instead of being ignored — remove stale parameters from persisted prompts). Rollback path: re-pin docker-compose.yml to 0.6.4, pip install jira-tempo-mcp==0.6.4, or npm i -g jira-tempo-mcp@0.6.4.
v0.6.4 — wheel-mode installer automation + report render fixes
Wheel-mode installer automation plus report render fixes. On a pip/wheel install (no git clone), jira-tempo-mcp install now fully configures VS Code itself — credentials prompts, mcp.json + .env.local registration, specialist install, idempotent re-runs — instead of erroring that a git clone is required.
Highlights
install/uninstallnow fully configure VS Code (mcp.json+.env.local) on pip/wheel installs without a git clone: credential prompts (PAT via hidden input), specialist install, idempotent.- Reports (txt and md) no longer re-print the issue summary echoed inside worklog comments.
- Tempo
/workers404 logs one warning instead of an error.
Verification
Full CI gate green: ruff, mypy, pytest — 505 passed (1 warning), wheel + sdist build.
Compatibility
No action needed. pip/wheel users can now re-run jira-tempo-mcp install to let the tool manage their VS Code config.
v0.6.3 — zcode/pi/hermes harnesses + specialist auto-refresh on update
This release extends the reporting-specialist installer to three more agent harnesses and makes the specialist maintain itself across updates. It also fixes a startup crash that affected fresh pip install since the mcp 2.x release.
Highlights
- Three new harnesses with confirmed conventions.
install-specialistnow supportszcode(agent file in~/.zcode/agents/plus skill dir in~/.zcode/skills/jira-tempo-reports/) andpi/hermes, both skills-only (~/.pi/agent/skills/jira-tempo-reports/and~/.hermes/skills/jira-tempo-reports/). The registry order is now copilot, zcode, claude, pi, hermes, opencode;codexremains skipped. - Specialist auto-refresh on update.
install-specialistrecords the harnesses it installed in a state file ($XDG_STATE_HOMEor~/.local/state/jira-tempo-mcp/specialist-state.json). After a successfuljira-tempo-mcp update, the specialist is re-installed from the new package data into every recorded harness, so reports stay in sync with the version you just installed. Per-harness failures warn and never change the exit code;install-specialist --removedeletes the state file. mcppinned to<2, fixing fresh-install breakage. mcp 2.x removed the low-levelServerdecorators (list_tools/call_tool) that the server relies on, so a freshpip install jira-tempo-mcpresolving mcp 2.x crashed at startup with anAttributeError. The dependency is now pinned to>=1.0.0,<2, and a drift-guard test (tests/test_mcp_compat.py) fails loudly if the API surface drifts again.
Verification
Full CI suite green: 467 tests passed, 1 warning; lint, type checks and package build all clean.
Compatibility
- Python installs: the
mcp<2pin resolves automatically; no action needed beyondpip install --upgrade jira-tempo-mcp. - Specialist migration: installs made before this release have no state file, so the new auto-refresh has nothing to act on yet. After updating, run
install-specialistonce — it re-installs into the harnesses you pick and records them, and every laterjira-tempo-mcp updaterefreshes them automatically. For the 0.6.2 → 0.6.3 upgrade specifically, this one manualinstall-specialistrun is required to activate auto-refresh.
jira-tempo-mcp v0.6.2
This release fixes a long-standing annoyance for anyone running both VS Code (Copilot) and Claude Code on the same machine: the specialist installer wrote the same agent file into two directories that VS Code scans, so the agent picker showed a duplicate "JTM: Jira Tempo Reports" entry. The Claude harness is now skills-only — Claude Code keeps working through its skill, and the picker shows a single entry.
Highlights
- Duplicate picker entry gone:
install-specialistno longer writes an agent file to~/.claude/agents/. Theclaudeharness now installs skills-only (the skill under~/.claude/skills/jira-tempo-reports/), which is all Claude Code ever used. VS Code no longer sees a second, identically-named file. - Cleaner harness directories: reinstall backups moved out of harness dirs into
~/.copilot/.backups/. Timestamped.bak.*files no longer accumulate next to live files in directories VS Code scans (a.bakwith malformed frontmatter could surface as a bogus picker entry via the filename fallback). - Removes legacy leftovers:
install-specialist --removenow purges remains of pre-skills-only installs —jtm-*files in~/.claude/agents/(including the file behind the duplicate entry) and stale in-tree.bak.*backups next to JTM targets in every harness dir. Foreign files are never touched.
Verification
make cigreen: ruff, mypy, pytest, build — 445 passed, 1 warning (unchanged suite; new tests cover the skills-only claude plan, backup relocation, and legacy purge).
Compatibility
- After upgrading, run once:
jira-tempo-mcp install-specialist --removefollowed byinstall-specialist. This removes the legacy Claude agents file that caused the duplicate VS Code picker entry and installs the new skills-only Claude layout. - No API or report-format changes; all MCP tools, report rendering, and templates are unchanged from v0.6.1.
- Rollback: pin the previous version (
pip install jira-tempo-mcp==0.6.1, npmjira-tempo-mcp@0.6.1, docker image:0.6.1).
v0.6.1 — worklog-dedup report fix
Weekly and team reports no longer show duplicated worklog blocks. When several entries on the same issue repeat the same row-header comment — typical per-day re-logging, or rows differing only by a typo in the detail line — they now render as a single block with the spent time summed (6h + 3h → 9h), keeping detail lines from all merged entries in the body. A fetch-level dedupe by Tempo worklog id additionally guards the weekly and team generators, logging a warning when it drops an entry; worklogs without an id are never dropped.
Highlights
- Reports stay readable after re-logs. Duplicated blocks collapse into one; the reported total now matches the actual logged time (6h + 3h → 9h) instead of double-counting.
- Nothing is lost in the merge. Detail lines present in only some of the merged entries are preserved in the final block.
- Protection against true duplicates. Fetch-level dedupe by Tempo worklog id (
tempoWorklogId) in weekly and team generators, with a warning log; id-less workloads are never dropped. - Self-update visibility. README quick-start (EN + RU) now shows
jira-tempo-mcp updateas the standard way to upgrade — pip upgrade for wheel installs, git pull + reinstall for editable installs.
Verification
Local release gate green from a clean build: make ci → lint (ruff) + typecheck (mypy) + 437 passed tests (1 warning) + build. Artefacts attached to this release: jira_tempo_mcp-0.6.1-py3-none-any.whl and jira_tempo_mcp-0.6.1.tar.gz (published to PyPI in the same pass).
Compatibility
No API or configuration changes — report formatting only. Upgrade from any 0.x version: pip install --upgrade jira-tempo-mcp (wheel installs), jira-tempo-mcp update, or npm i -g jira-tempo-mcp (npm wrapper).
v0.6.0 — self-update, install-specialist, npm wrapper
jira-tempo-mcp 0.6.0 teaches the server to maintain itself: a new update command keeps your install current, a new install-specialist command drops a ready-made Jira/Tempo agent into your AI harness, and an npm wrapper lets Node-centric setups install everything with a single npm i -g jira-tempo-mcp.
Highlights
- Keep your install up to date with one command.
jira-tempo-mcp updatefigures out how the tool was installed — a git checkout, a pip install from PyPI, or a Docker/PYTHONPATHrun — and updates accordingly or tells you plainly what to do by hand. A--checkmode shows whether an update is available without touching anything. Updates always land in the environment you are actually running. - Install a preconfigured Jira/Tempo agent into your harness.
jira-tempo-mcp install-specialistcopies thejtm-jira-tempo-reportsagent and skill files into Copilot, Claude Code, or OpenCode (the harness table is in the docs). Codex is intentionally not supported — it has no agent-file convention — and the command says so explicitly rather than writing somewhere unexpected. Works from a plainpip install, no git clone needed. - One-line install for npm users. The new npm wrapper package installs the Python server from PyPI and provides a
jira-tempo-mcpcommand on your PATH (Node ≥ 18). Useful when your harness or workflow pulls tools from npm. - Docs refreshed to match. The CLI reference now covers
updateandinstall-specialist; the installation guide documents the PyPI/npm paths with a harness-coverage table; README quick-links updated.
Verification
The release gate ran on a fresh build: make ci — lint, typecheck, 426 tests passed, and a clean build of the 0.6.0 wheel and source distribution. All green.
Compatibility
No breaking changes: all 19 existing MCP tools, the config schema, and existing install paths behave exactly as in 0.5.0. Rolling back is a standard downgrade — pip install jira-tempo-mcp==0.5.0 (or re-pin your previous Docker image). The npm wrapper is a new package version; installing 0.5.0 from npm is not possible, so npm users wanting to roll back should uninstall the wrapper and install the Python package directly.
v0.5.0 — issue description access, Jira write tools, task templates
What's new
jira-tempo-mcp v0.5.0 adds read access to Jira issue descriptions, two write tools (issue creation and commenting), and a task-template system that creates an issue with a full set of subtasks from a YAML template. The server now covers 19 MCP tools (previously 17).
Highlights
- Issue descriptions in search results and single-issue reads —
get_issuenow returns the issue description alongside the other fields, andlist_issues_by_jqlaccepts an optionalinclude_descriptionflag (off by default, so list responses stay compact). - Create Jira issues from the chat — the new
create_issuetool creates an issue in a given project with summary, optional description, issue type, and optional parent (for subtasks or epic links). - Comment on issues — the new
add_issue_commenttool posts a comment to an issue and rejects empty comments on both the tool and API side. - Create an issue with subtasks from a template —
create_issue_from_templatetakes a YAML task template and creates the parent issue plus its child subtasks in order, stopping on the first failure and reporting what was created so far.list_issue_templatesshows what's available. - Built-in stand-preparation template — ships with 15 environment-provisioning child tasks. Custom templates load from a directory configured with
JTM_TEMPLATES_DIR, and a file there with the same name overrides the built-in.
Verification
- 391 tests passing (up from 328),
make cigreen (ruff, mypy, pytest, package build). - Deep code review completed with zero blocking findings.
- Read-only end-to-end verification against a live Jira instance: all 19 tools present, template creation path and description exposure confirmed.
Compatibility
No breaking changes: all existing tools and configuration keep their contracts. The new include_description option defaults to false. Docker image: ghcr.io/korrnals/jira-tempo-mcp:0.5.0 (and :latest).
v0.4.3
[0.4.3] — 2026-08-08
Fixed
jira-tempo-mcp installnow shows clear guidance when invoked outside a git clone (wheel/Docker). Previously failed with terse "install.py not found". install.py is a dev-setup script requiring the repo tree (.env.example,copilot-integration/,pyproject.toml); it is intentionally NOT shipped in the wheel.- Added tests covering the install-guidance message for non-clone installs.
Docs
installation.md/installation.ru.md: clarified thatjira-tempo-mcp installrequires git clone (editable install); Docker runs without install.