v0.5.0

Added

• Explorations (Metabase 0.65+, every tier, no AI provider needed): list_exploration_dimensions,
  create_exploration, get_exploration, get_exploration_query_result, list_my_explorations,
  update_exploration, delete_exploration, restart_exploration_thread, cancel_exploration_thread.
  create_exploration takes metric card IDs and dimension IDs, sends each metric with its dimension
  mappings (without them Metabase plans no queries), and waits for the queries to finish.
• New toolset explorations (in the query preset).

Changed

• API types regenerated from the Metabase 0.65 spec; CI pinned to v0.65.2.6.

v0.4.0

Breaking

• Query tools return compact output by default; pass verbose: true for the raw Metabase response.
• list_tasks moved from the activity toolset to diagnostics.

Large-result safety

• Query tools (execute_query, execute_card_query, run_metric) return a compact view by default:
  columns (name, type) + rows, no column internals, results_metadata or insights. verbose: true
  returns the raw Metabase response.
• Row cap METABASE_MAX_ROWS (default 500) with per-call max_rows and offset paging; a note
  says how to get the rest (next offset, or an export tool with output_path).
• Every tool response is serialised as compact JSON and capped at METABASE_MAX_RESPONSE_CHARS
  (default 100000) with a truncation notice.
• Measured on a 10 / 100 / 500-row SELECT * FROM ORDERS: 94% / 76% / 60% fewer characters than 0.3.0.

Resources and prompts

• MCP resources: metabase://databases, metabase://database/{id}/schema, metabase://card/{id},
  metabase://dashboard/{id}, metabase://collection/{id} (compact JSON summaries).
• Prompts: explore-database, build-dashboard, explain-card, audit-permissions.
• Both follow the tool selection: resources need their toolset, prompts need every tool they use.

Admin tools (17 new, 178 total)

• API keys toolset: list_api_keys, create_api_key (full key returned once),
  update_api_key, regenerate_api_key, delete_api_key.
• Embedding toolset: get_embedding_status (settings, embeddable items, whether a signing key
  exists, never its value), set_static_embedding, set_resource_embedding (per-parameter
  enabled/locked/disabled), create_embed_url (signed iframe URL, signed locally).
• Tasks & logs (diagnostics) toolset: list_tasks (now with status/task filters, paging and
  compact output; moved from activity), get_task, list_task_types, list_task_runs,
  get_scheduler_info, get_server_logs (level/text filter, ANSI stripped), list_log_presets,
  set_log_levels / reset_log_levels (temporary), get_premium_token_status.
• The admin preset includes the three new toolsets.

Types

• generated/metabase-api.d.ts regenerated from the Metabase v0.64 OpenAPI spec (the CI version;
  scripts/generate-types.ts is now pinned to it, METABASE_SPEC overrides). 27 as any casts
  removed where the new types cover the calls.

v0.3.0

Compatibility release for Metabase 0.60–0.64 (verified on 0.59.6 and 0.64.3.4) and a headless-first rework.

Highlights

• Fixes for Metabase 0.60+ API changes (card parameters, segments, schema-less DBs) and ~10 tools that never worked (alerts → /api/notification, dashboard subscriptions, persistence, activity, glossary, documents, EID translation, CSV upload, collection sorting).
• 161 tools (was 128): metrics, measures, transforms, exports (CSV/JSON/XLSX, dashboard PDF), notifications, get_group.
• Headless HTTP mode (MCP_TRANSPORT=http) with bearer token; toolsets and per-tool enable/disable; readOnly/destructive hints on every tool.
• Ships everywhere: npm, official MCP Registry (io.github.Serdnaley/metabase-mcp), Claude Desktop .mcpb bundle (attached below), Docker ghcr.io/serdnaley/metabase-mcp.
• MIT license, tool catalog: https://serdnaley.github.io/metabase-mcp/

Breaking: create_alert / update_alert take a new input shape (the old one never worked); notify_database removed (use sync_database_schema).

Full list: CHANGELOG.md

v0.2.0

128 MCP tools across 26 domains

Major expansion from 42 to 128 tools covering the full Metabase REST API surface.

New domains (18)

• Users & Groups (13 tools) — full lifecycle, group membership
• Permissions (4) — data and collection permission graphs
• Segments (5) — reusable named filters CRUD
• Database write/sync (6) — connections, validation, sync triggers
• Notifications (9) — alerts, dashboard subscriptions
• Revisions (2) — history and revert
• Settings (3) — instance configuration
• Glossary (5) — business term definitions
• Documents (5) — narrative analytics
• Timelines (8) — chart annotations
• Bookmarks (3), Moderation (1), EID Translation (1)
• SQL Snippets (5), Activity/Tasks (3), X-Rays (1)
• Cache (2), Model Persistence (3), CSV Upload (1)
• Action Public Links (2)

Enhanced existing tools

• Cards: type (question/model/metric), result_metadata, enable_embedding
• Tables: write ops (update_table, update_field, rescan_field_values, discard_field_values, reorder_fields)
• Collections: get_collection_tree, personal_only + archived filters
• Search: archived filter
• Segments: table_id filter

Testing

• 202 tests, 668 assertions across 31 test files

See CHANGELOG.md for full details.

v0.1.5

What's Changed

• feat: add public link tools, fix date_range params, improve visualization settings descriptions by @Serdnaley in #1
• chore: prepare 0.1.4 release — changelog, README updates for public link tools by @Serdnaley in #2
• chore: bump to 0.1.5 by @Serdnaley in #3

New Contributors

• @Serdnaley made their first contribution in #1

Full Changelog: v0.1.4...v0.1.5

v0.1.4

CI fix: repository URL case for provenance

v0.1.3

CI fix: Trusted Publishing with Node 24

v0.1.2

Added

• CLAUDE.md for Claude Code context
• CHANGELOG.md

See full changelog

v0.1.1

Fixed

• execute_card_query now supports parameterized cards with template tags (e.g. fromDate, toDate). Pass values via parameter_values map.

Full changelog

See CHANGELOG.md