Fixes a false configured: false from the get_setup_status MCP tool when the server runs in legacy inline mode (the launch shape the Claude Code plugin UI uses: --host/--user/--dbname + REDSHIFT_PASSWORD env). The status tool now detects inline mode and adds a source: "inline" | "profile" field so agents can tell which config mechanism is live. The password value is never materialized in the response (presence-only bool).

See #39.

Pre-release: triggers TestPyPI smoke test. PyPI-live promotion follows after the dry-run goes green.

What's Changed

• feat(plugin): connection-config userConfig form (reverse D2) by @kouko in #36
• feat(plugin): default-disabled install (opt-in connector) by @kouko in #37
• feat(mcpb): companion Desktop Extension — Claude Desktop config form by @kouko in #38

Full Changelog: v0.7.1...v0.9.0

🎯 In-band MCP setup — no log-spelunking, no restart

v0.7.1 is the catch-up release that ships v0.7.0's work to PyPI. The
v0.7.0 Pre-release attempt was blocked by two GitHub Actions workflow
quirks (TestPyPI CDN propagation lag + non-idempotent twine upload) that
made test-publish.yml fail spuriously. v0.7.0 was tagged on git and
uploaded to TestPyPI, but never reached PyPI — bumping to 0.7.1 with the
workflow fix (PR #35) is the cleanest path to a green ship, mirroring
the v0.4.0/v0.5.0 → v0.6.0 catch-up pattern.

v0.7.1 = v0.7.0 + workflow fix — functionally identical to v0.7.0
from the user perspective.

agent: list_schemas() → {"error": "not_configured", "exception_class": "ConfigurationError", "next_step": "Call setup_via_dialog..."}
agent: setup_via_dialog(host=..., user=..., dbname=...)
       → OS dialog appears, user types password
       → SELECT 1 against Redshift to verify
       → {"status": "configured", "tested": true}
agent: list_schemas() → works (lazy resolve, no restart)

✨ New MCP tools (11 → 13)

🆕 Degraded-mode startup

Pre-v0.7.x: server crashed with ValueError("Profile 'default' is not configured...") if no profile existed. Agents had no in-band signal — they had to read ~/Library/Logs/Claude/mcp-server-*.log (client-specific path) to find the recovery hint.

v0.7.x:

• Server boots regardless of profile state — enters the MCP stdio loop and accepts tool calls
• DB tools return a structured {"error": "not_configured", ...} response with next_step field pointing at setup_via_dialog
• Lazy connection resolution: every tool call re-reads config.toml + keychain, so newly written profiles take effect without restarting the MCP client
• resolve_connection_params now raises ConfigurationError (a ValueError subclass for backward-compat)

🔐 Password collection: OS dialog, never the wire

The same osascript / zenity discipline already used by the Claude Code /redshift-setup skill is now in the CLI + MCP tool:

Password value path: user keyboard → OS dialog process stdout (in-pipe) → Python string in-process → keyring.set_password() → OS keychain. Never touches MCP wire, chat, Bash stdout, process args, or shell history.

🛡️ macOS Apple Events permission detection

v0.7.x distinguishes 3 macOS-specific osascript failure modes that pre-v0.7.x all collapsed to "user cancelled":

Detection uses the locale-stable numeric code (-1743) plus case-insensitive English text fallback ("Not authorized" / "Not authorised").

📋 Unified error-response schema

All four "this is broken" responses now carry the same field set so agents can pattern-match a single shape:

{
  "error": "not_configured" | "write_profile_failed" | "keychain_write_failed" | "missing_field",
  "exception_class": "ConfigurationError" | "PermissionError" | "KeyringLocked" | "ValidationError" | ...,
  "message": "<sanitized agent-actionable text>"
}

Per CWE-209 + OWASP Error Handling Cheat Sheet, raw str(exc) is not included for write_profile_failed / keychain_write_failed (third-party exception args may carry sensitive context). The exception class name is preserved as a separate field for diagnostic branching. Full exception detail is logged server-side with exc_info=True.

⚡ Connection test before declaring success

setup_via_dialog now runs SELECT 1 against Redshift after writing the profile + password. Catches host typos / VPN-not-connected / wrong password / firewall block / paused-cluster early — agent gets either {"status": "configured", "tested": true} or {"status": "configured_but_connection_failed", "connection_error": "..."} instead of a misleading "configured" followed by an immediate DB tool failure.

Failures are also logged server-side at WARNING level with profile + cluster coords + the underlying error, so operators can correlate transcript-reported issues with server log entries.

🤖 FastMCP instructions block updated

The handshake-time instructions= text now describes:

• The degraded-mode contract (server boots without profile)
• Both entry points: proactive get_setup_status at session start vs reactive not_configured on tool call
• All response statuses agents need to handle, including permission_denied and configured_but_connection_failed

Agents see the recovery pattern BEFORE encountering an error, not just inside the error itself.

📚 Trilingual README updates

README.md / README.ja.md / README.zh-TW.md all updated with:

• Tool count 11 → 13 + new setup_via_dialog and get_setup_status rows
• Code-agent bootstrap diagram showing the v0.7.x pure-MCP flow
• Fallback section for headless / no-GUI hosts (--stdin pipe path)

🧪 Test coverage

Highlights:

• test_bootstrap_then_use_end_to_end — the keystone scenario: server boots without profile → agent calls setup_via_dialog → next DB tool call works without restart
• test_server_main_does_not_crash_when_no_profile — regression net against re-introducing upfront resolution
• test_every_db_tool_handles_not_configured — parametric over all 11 DB tools (catches missing @_guarded on new tools)
• test_get_setup_status_never_returns_password_value — hard security invariant; JSON-serializes response and scans for the password substring
• test_collect_password_via_dialog_detects_permission_denied — parametric over realistic macOS stderr shapes (numeric, English, British spelling, lowercase, locale-translated)

🔧 Refactor + dev experience

• RedshiftTools.__init__ now takes a Callable[[], RedshiftConnectionConfig] provider instead of a built config — enables lazy re-resolution. 58 existing RedshiftTools(config) test call sites mechanically rewritten to RedshiftTools(lambda: config).
• @_guarded decorator on all 11 DB tools centralizes the ConfigurationError → not_configured response conversion. Two new tools (setup_via_dialog, get_setup_status) deliberately skip the guard since they must run when no profile exists; documented inline + behaviorally tested.
• 5 dialog-failure response builders extracted to module-level functions with a _DIALOG_FAILURE_BUILDERS dispatch dict — setup_via_dialog body shrinks from ~250 → ~164 lines.

🛠️ Release workflow fixes (v0.7.1 only — PR #35)

Two GitHub Actions workflow flaws surfaced by the v0.7.0 release attempt, fixed in v0.7.1:

1. test-publish.yml "Test installation from TestPyPI" step now retries pip install up to 6× with 30s backoff (3 min total) — absorbs the typical 1-3 min lag between TestPyPI's JSON API and its simple index (which pip uses).
2. twine upload now uses --skip-existing in both test-publish.yml and publish.yml — re-runs are idempotent (no more 400 Bad Request on duplicate upload). Before this fix, any transient failure during the publish workflow corrupted the version slot until the next bump.

Forward-looking: future releases on the same workflows will not require version skipping just to retry.

📥 Install

# PyPI / generic MCP client
pip install --upgrade redshift-comment-mcp
# or
uvx redshift-comment-mcp

# Claude Code plugin
claude plugin install redshift-comment-mcp

pip install --upgrade from v0.6.0 lands directly on v0.7.1 (v0.7.0 is not on PyPI; see workflow fixes section above).

For multi-cluster claude_desktop_config.json / mcp.json examples and the in-band setup_via_dialog agent flow, see README §"Setting up with uvx".

🔗 Forward look: MCP Elicitation primitive

The MCP 2026-07-28 RC spec introduces an Elicitation primitive — a protocol-level mechanism for servers to request direct user input through the client's native UI. Once GA and FastMCP exposes it, the subprocess + osascript dialog hack here becomes obsolete: the server can request a password via the protocol and the client presents a native secure input widget. Documented in code at redshift_tools.py:1122-1140 so future maintainers know this is interim, not permanent architecture.

📊 External references grounding this release

• [CWE-2...

⚠️ Superseded — never reached PyPI, use v0.7.1 instead

This tag exists in git history and on GitHub Releases, but was never
published to PyPI. The v0.7.0 release attempt was blocked by two
GitHub Actions workflow flaws:

1. test-publish.yml's "Test installation from TestPyPI" step ran ~30s
   after upload — TestPyPI's pip-facing simple index typically lags the
   JSON API by 1-3 minutes, so pip couldn't see 0.7.0 yet and the step
   failed spuriously.
2. The twine upload step wasn't idempotent (--skip-existing missing),
   so gh run rerun --failed hit HTTPError: 400 Bad Request from
   the duplicate-version check.

The first failure was a cosmetic timing issue (TestPyPI actually had
0.7.0); the second made the workflow status permanently stuck on
"failed" with no clean recovery path. Bumping to 0.7.1 + fixing the
workflows (PR #35) was the cleanest path to a green ship — same
pattern as v0.4.0/v0.5.0 → v0.6.0.

Migration: install v0.7.1
or later. v0.7.1 is functionally identical to v0.7.0 from the user
perspective — only the publish workflows differ.

pip install --upgrade redshift-comment-mcp     # → 0.7.1+
# or
uvx redshift-comment-mcp                       # → 0.7.1+

The original v0.7.0 release notes are below for historical reference.

🎯 In-band MCP setup — no log-spelunking, no restart

v0.7.0 collapses the "code-agent provisions a Redshift profile" UX from
7 steps (with log-spelunking + Bash + MCP-client restart) down to
3 steps, pure MCP wire, no Bash needed. The server now boots even
without a configured profile and exposes a dedicated setup_via_dialog
tool that drives an OS-native password dialog server-side — password
never crosses chat / MCP wire / tool args.

agent: list_schemas() → {"error": "not_configured", "exception_class": "ConfigurationError", "next_step": "Call setup_via_dialog..."}
agent: setup_via_dialog(host=..., user=..., dbname=...)
       → OS dialog appears, user types password
       → SELECT 1 against Redshift to verify
       → {"status": "configured", "tested": true}
agent: list_schemas() → works (lazy resolve, no restart)

(Full v0.7.0 changelog identical to v0.7.1's notes minus the §"Release workflow fixes" section that's unique to v0.7.1.)

🎯 PyPI back in sync with the plugin path

First PyPI release since v0.3.0. The Claude Code plugin path shipped
v0.4.0 + v0.5.0 internally from cloned-repo source, but neither was
published to PyPI due to a release-trigger workflow bug — v0.6.0 rolls
up both prior releases plus the uvx UX work that promotes
uvx redshift-comment-mcp to a first-class install path for non-Claude-
Code MCP clients.

The new Release publishing (MUST) discipline (#31) + publish.yml
trigger fix (#33) close the gap that caused the silent v0.5.0
non-publication. v0.5.0 is intentionally skipped on PyPI; v0.6.0 is
its strict superset plus the UX work.

🤖 New in v0.6.0 — code-agent uvx bootstrap (#33)

uvx redshift-comment-mcp becomes truly stand-alone — no Claude Code
plugin required for first-time setup, and the password stays out of
chat / stdout via OS-native dialog.

server.py's "profile not configured" error messages now offer three
setup paths: Claude Code skill, code-agent pipeline (set-fields +
set-password --dialog), and human terminal (uvx redshift-comment-mcp setup). Before v0.6.0, the error only mentioned the Claude Code skill —
uvx-only users hit a dead-end message.

Trilingual README §"Setting up with uvx" now ships concrete
claude_desktop_config.json snippets (single-profile + multi-cluster)
plus the code-agent bootstrap pipeline (#32 + #33).

✨ New skills (rolled up from v0.4.0)

🆕 Other new features (v0.4.0 + v0.5.0)

• execute_sql transparency — response now includes _executed_sql
  (the rewritten SQL the server actually ran) and _user_facing_message
  (a shaped string the agent can show users) (#28).
• Zero-config plugin install — claude plugin install no longer
  asks for a profile name. The MCP server reads an active-profile
  pointer file written by /redshift-setup (#21, breaking, see
  migration below).

⚡ Performance (rolled up from v0.4.0)

• Column metadata SQL: ~3.3x at 12K columns. Rewrote correlated-
  subquery pattern to direct LEFT JOIN. Speedup persists at small N (#22).
• Long-comment safety in multi-item responses. MAX_COMMENT_LEN=1000
  caps each comment with a comment_truncated_count marker; single-item
  getters never truncate (#23).

🐛 Bug fixes

• Single-profile upgrade rescue (v0.5.0) — pre-PR-22 installs whose
  lone profile was not named default got hard-broken by the active-
  profile pointer file introduction. resolve_active_profile now falls
  back to the lone profile when exactly one exists; explicit
  --profile / env / pointer inputs still resolve verbatim so typos
  surface as typos (#29).
• uvx-only error UX (v0.6.0) — "no profile configured" error now
  guides ALL setup paths, not just the Claude Code skill (#33).
• Various skill bug fixes from functional-test pass (#19).

🏠 Housekeeping

• Removed /redshift-cache-schema — cache layer retired in favor
  of always-fresh catalog SQL + the new performance work (#26).
• Live-cluster smoke gate. REDSHIFT_INTEGRATION=1 pytest tests/integration/
  exercises every MCP tool against a real Redshift cluster (#25).
• New tests/e2e/ tier — MCP wire-protocol tests via stdio
  subprocess (#28).
• README install-link fix (#27).

⚠️ Breaking — Profile system contract (rolled up from v0.4.0, #21)

If you installed v0.3.0 directly via pip install redshift-comment-mcp:

• Profile pointer file is new. ~/.config/redshift-comment-mcp/active-profile
  is now the source of truth for the active profile. Absence is no longer
  an error — single-profile users get an implicit fallback (after #29).
• Plugin manifest no longer asks for profile name. Anyone who
  manually set a profile name via client config should switch to
  --profile <name> to override the pointer file, or rely on
  /redshift-setup + /redshift-switch-profile.

If your existing single profile is named anything other than default,
upgrading straight to 0.6.0 (not stopping at any 0.4.x / 0.5.0 point)
is the safe path — the single-profile fallback rescue (#29) is included.

🛠️ Release discipline (v0.6.0 internal)

• CLAUDE.md now codifies the release-publishing flow as a hard
  discipline (tag → Pre-release → flip-to-Release → PyPI), so the silent
  non-publication that hit v0.4.0 + v0.5.0 doesn't recur (#31).
• .github/workflows/publish.yml trigger gains released activity
  type, so gh release edit --prerelease=false (the Pre-release →
  Release promotion) actually fires the workflow. v0.5.0 was the worked
  example of this trigger gap — fixed forward in v0.6.0 (#33).

📥 Install

# PyPI / generic MCP client
pip install --upgrade redshift-comment-mcp
# or
uvx redshift-comment-mcp

# Claude Code plugin
claude plugin install redshift-comment-mcp

📊 Full changelog

PRs in this release (v0.3.0…v0.6.0):
#19 · #20 · #21 · #22 · #23 · #24 · #25 · #26 · #27 · #28 · #29 · #30 ·
#31 · #32 · #33

⚠️ Never reached PyPI — use v0.6.0 instead

This tag exists in git history and on GitHub Releases, but was never
published to PyPI. A publish.yml workflow trigger bug (gh release edit --prerelease=false fires the released event, not published)
caused the Pre-release → Release promotion to silently miss the publish
job. PyPI users went straight from v0.3.0 to v0.6.0.

Migration: install v0.6.0
or later. v0.6.0 is a strict superset of v0.5.0 and ships the trigger
fix forward so this can't recur.

pip install --upgrade redshift-comment-mcp     # → 0.6.0+
# or
uvx redshift-comment-mcp                       # → 0.6.0+

The original v0.5.0 release notes are below for historical reference.

🎯 PyPI back in sync with the plugin path

This is the first PyPI release since v0.3.0. The Claude Code plugin path
(claude plugin install, runs from cloned-repo source) shipped v0.4.0 +
v0.5.0 internally, but neither was tagged or released — so PyPI /
uvx redshift-comment-mcp users were stuck on v0.3.0 with no visible
signal. v0.5.0 rolls up both releases' worth of work plus a critical
post-release patch.

Per the new Release publishing (MUST) discipline (#31), this won't
happen again.

✨ New skills

🆕 New features

• execute_sql transparency — response now includes _executed_sql
  (the actual rewritten SQL the server ran) and _user_facing_message
  (a shaped string the agent can show users), so the user always knows
  exactly what query hit Redshift (#28)
• Zero-config plugin install — claude plugin install no longer asks
  for a profile name. The MCP server reads an active-profile pointer file
  written by /redshift-setup. One-time /redshift-setup → done
  (#21, breaking, see migration below)

⚡ Performance

• Column metadata SQL: ~3.3x at 12K columns. Rewrote the correlated-
  subquery pattern to a direct LEFT JOIN. Speedup persists at small N (#22)
• Long-comment safety in multi-item responses. MAX_COMMENT_LEN=1000
  caps each comment with a comment_truncated_count marker; single-item
  getters never truncate. Prevents Claude Code's silent ~25K token cap
  from clipping dbt-rich schemas (#23)

🐛 Bug fixes

• Single-profile upgrade rescue. PR #22 introduced the active-profile
  pointer file with the contract "absent ↔ use default". Pre-PR-22
  installs whose lone profile was not named default (e.g.
  ichef-prod) got hard-broken by the upgrade. resolve_active_profile
  now falls back to the lone profile when exactly one exists; explicit
  --profile / env var / pointer file inputs still resolve verbatim so
  typos surface as typos (#29)
• Various skill bug fixes from functional-test pass (#19)

🏠 Housekeeping

• Removed /redshift-cache-schema — cache layer retired in favor of
  always-fresh catalog SQL + the new performance work above (#26)
• Live-cluster smoke gate. REDSHIFT_INTEGRATION=1 pytest tests/integration/
  exercises every MCP tool against a real Redshift cluster (#25)
• New tests/e2e/ tier — MCP wire-protocol tests via stdio subprocess (#28)
• README install-link fix (#27)

⚠️ Breaking — Profile system contract (#21)

If you installed v0.3.0 directly via pip install redshift-comment-mcp:

• Profile pointer file is new. ~/.config/redshift-comment-mcp/active-profile
  is now the source of truth for the active profile. Absence is no longer
  an error — single-profile users get an implicit fallback (after #29).
• Plugin manifest no longer asks for profile name. Anyone who manually
  set a profile name via client config should switch to --profile <name>
  to override the pointer file, or rely on /redshift-setup + /redshift-switch-profile.

If your existing single profile is named anything other than default,
upgrading straight to 0.5.0 (not stopping at any 0.4.x point) is the
safe path — the single-profile fallback rescue (#29) is included.

📥 Install

# PyPI / generic MCP client
pip install --upgrade redshift-comment-mcp
# or
uvx redshift-comment-mcp

# Claude Code plugin
claude plugin install redshift-comment-mcp

📊 Full changelog

PRs in this release (v0.3.0…v0.5.0): #19 · #20 · #21 · #22 · #23 · #24 ·
#25 · #26 · #27 · #28 · #29 · #30

⚠️ Breaking changes

Two skills are removed in this release. Charter narrowed: skills exist
to help users do natural-language search and analysis of data;
artifact generation falls outside that scope.

Final skill set (5): /redshift-setup, /redshift-explore,
/redshift-profile, /redshift-cache-schema, /redshift-lineage-from-stl.

🚀 Cache layer repurposed as LLM-internal lookup

/redshift-cache-schema switches role from human offline-browsing
artifact to an LLM-side metadata cache that other skills actually
consume.

• New file format: per-table tables/<schema>__<table>.md preserving
  multi-line markdown comments verbatim, plus lossy TSV indices
  (_tables_index.tsv / _columns_index.tsv) for cross-cluster grep,
  plus _meta.json freshness gate (refreshed_at, ttl_hours,
  complete: true|false).
• Server instructions teach CACHE PROTOCOL: agents prefer Reading
  cache files over MCP round-trips when complete=true and within TTL,
  fall back to live MCP on miss with a chat hint to refresh.
• /redshift-explore Step 0 + /redshift-profile Step 1 hook into the
  cache-first path. Other consumer skills benefit automatically via the
  server-level CACHE PROTOCOL.

Expected impact when cache is fresh: ~700 tokens saved + 99% latency
reduction per metadata query (Bash grep on TSV vs search_* MCP, or
single-file Read vs list_columns(include_comments=True)).

🔧 Metadata trim (system prompt — every conversation)

Achieved via description-design.md principles
(WHAT + WHEN, not WHAT + WORKFLOW; keyword-belt multi-language triggers
instead of full sentences; no all-caps emphasis density;
deduplicated comment-authoritative rule from 4 places to 1).

📝 Known limits documented

Three-language READMEs now document Claude Code's
~25K token MCP response cap
(silent truncation, no error marker), the two mitigations this plugin
already applies (include_comments=False default + cache layer), and
the MAX_MCP_OUTPUT_TOKENS env-var override.

✅ Test protection: 98 → 142 tests

Three new test files catch the classes of bug encountered during this
refactor cycle:

All pure-function tests; run in <100 ms; require no AWS / MCP / DB.

🔐 SQL execute_sql validator (already in v0.2.2, unchanged here)

The validator changes from v0.2.2 (BOM-safe, string/comment-aware,
unterminated-literal rejection, 90.48% coverage gated at 85%) ship as-is
in this release.

📊 Changes Summary

🔄 Upgrade

pip install --upgrade redshift-comment-mcp
# or
pip install redshift-comment-mcp==0.3.0

If you used /redshift-erd or /redshift-suggest-schema-yml, see the
migration notes above.

Full Changelog: v0.2.2...v0.3.0

🤖 Generated with Claude Code

🚀 Claude Code Plugin Support

This release makes redshift-comment-mcp installable as a Claude Code plugin
with a one-shot conversational onboarding flow — no manual MCP server config
required.

• /redshift-setup slash command (PR #9, refined in PR #13) — 4-question
  conversational flow that configures a connection profile end-to-end. Default
  form silently uses the default profile name; /redshift-setup <name> opts
  into multi-cluster mode and auto-writes ~/.claude/settings.json (with
  BEFORE/AFTER preview)
• Setup CLI redshift-comment-mcp setup (PR #7) — same flow without the
  conversational layer, for non-Claude-Code users
• OS-aware password input (PR #9) — system password dialog when GUI is
  available, TTY fallback otherwise
• 6 MCP-composed skills + tri-lingual docs (EN / JP / 繁中) (PR #9):
  /redshift-profile, /redshift-suggest-schema-yml,
  /redshift-cache-schema, /redshift-erd, /redshift-explore,
  /redshift-lineage-from-stl
• Plugin runs from cloned repo source via uv run --project (PR #7) —
  no PyPI release required for plugin users (PyPI install path still supported)

🔐 SQL Security Hardening

execute_sql now stands up to a wider class of payloads. Two fixes shipped:

Expanded blocklist (PR #11). Added MERGE, GRANT, REVOKE, COPY,
UNLOAD to the forbidden-keyword list — covers Redshift upsert, ACL
mutation, and S3 import / export (data exfiltration surface even though
the latter doesn't "write to DB" in the strict sense).

Robust validator (PR #14). Reproduces a real failure where a valid
WITH ... SELECT query was rejected because of a transport-layer BOM
prefix breaking str.strip().startswith('WITH'). Now:

29 new parametrized tests covering BOM / ZWSP / NBSP prefixes, string &
comment false-positives, multi-statement piggyback, and unterminated
literals. Validator module coverage at 90.48%, locked behind a CI gate
that fails the build below 85%.

🛠 Build & CI Robustness

• fallback_version for plugin cache installs (PR #12) — Claude Code
  clones plugins without .git, so setuptools_scm had nothing to read at
  install time. Falls back to a static version string in that case
• FastMCP 2.x ↔ 3.x compat in test helper (PR #8, refined in PR #14) —
  _get_tool_fn now tries public list_tools() first, falls back to private
  _list_tools. Suite green on both lines
• Pin fastmcp>=2.11.0 and add pytest-cov to dev deps (PR #14)
• Coverage gate on redshift_tools.py (≥85%, currently 90.48%) (PR #14)
  blocks regressions in security-critical code
• Stale comment fix (PR #10) — removed inaccurate claim that startup
  performs a connection test

📊 Changes Summary

✅ Test Coverage

• 98 tests passing (was 52 in v0.2.0)
• 90.48% coverage on redshift_tools.py, locked at 85% via CI gate
• Python 3.10 / 3.11 / 3.12 matrix in CI

Full Changelog: v0.2.0...v0.2.2

🤖 Generated with Claude Code

🚀 New Features

Search Tools Enhancements

• search_schemas - New tool to search schemas by keywords (name or comment)
• hit_count ranking - Search results now prioritize items matching more keywords
  • Results sorted by hit_count DESC, name ASC
  • hit_count field included in response

List Tools Optimization

• include_parent_comments parameter - Control whether to fetch parent-level comments
  • list_tables: skip schema comment query when False
  • list_columns: skip table comment query when False
• include_comments default changed to False for list_tables and list_columns
  • Reduces unnecessary SQL queries
  • list_schemas keeps default True

MCP Instructions Updated

• Added search_schemas to recommended exploration flow
• Emphasized keyword language should match user's conversation language

📊 Changes Summary

✅ Test Coverage

• 52 tests passing
• New tests for hit_count sorting (3 tests)
• New tests for include_parent_comments parameter

🤖 Generated with Claude Code

🔧 修正與改進

Bug 修正

• SQL 關鍵字驗證修正：使用 \b regex 字元邊界檢查危險 SQL 關鍵字，避免誤擋含有關鍵字的欄位名稱（如 last_update）

文件更新

• README 強化：新增完整的 MCP Client 本地開發設定說明
  • 強調 Python 完整路徑的重要性
  • 新增環境變數隱藏密碼的設定範例
  • 說明更新原始碼後的操作流程

其他

• 新增 _version.py 到 .gitignore（setuptools-scm 自動生成檔案）

📥 安裝

pip install redshift-comment-mcp==0.1.1

🔄 升級

pip install --upgrade redshift-comment-mcp