Skip to content

Releases

safe-db v1.1.1

Choose a tag to compare

@COBOL-Kid COBOL-Kid released this 18 Nov 04:31
5254ead

1.1.1

Desktop app, MCP server (safe-db-mcp / @safe-db/mcp), and shared query engine. Version is 1.1.1 in build.gradle.kts; the MCP fat JAR is safe-db-mcp-1.1.1-all.jar; npm packages are stamped 1.1.1 at assemble time.

  • Settings query timeout is configurable from 1 to 120 seconds, default 10; it applies to the next desktop or MCP query run and its plan check; connect and schema introspection stay fixed. safe-db-mcp settings prints query_timeout_seconds; safe-db-mcp settings timeout writes it.
  • The sidebar shows a reachability indicator (Connected / Not connected) for the selected connection, and the Connections screen labels that connection Selected instead of Active.
  • Updated the bundled Temurin 25 jlink runtime for @safe-db/mcp to jdk-25.0.5+8 and updated runtime dependencies (mssql-jdbc 13.6.2, kotlinx-serialization-json 1.13.0) (#112).
  • Refreshed README and documentation links to safe-db.dev and trimmed "single SELECT" wording from MCP tool descriptions and CTE error copy (#110).

safe-db v1.1.0

Choose a tag to compare

@COBOL-Kid COBOL-Kid released this 14 Nov 04:49
9ecac84

1.1.0

Desktop app, MCP server (safe-db-mcp / @safe-db/mcp), and shared query engine. Version is 1.1.0 in build.gradle.kts; the MCP fat JAR is safe-db-mcp-1.1.0-all.jar; npm packages are stamped 1.1.0 at assemble time.

  • Schema discovery and join guidance: describe_table includes join_eligible on columns and incoming_foreign_keys. Added search_catalog for case-insensitive matching across tables, columns, indexes, and foreign keys, and find_join_paths for direct neighbors or multi-hop foreign-key and indexed-column paths.
  • Dry-run query assessment: added assess_query (SafeDbService.assessQuery / assessQueryCore) to parse, validate, compile, run EXPLAIN, and evaluate the risk gate and confirmation requirements without executing the SELECT or writing query history.
  • In-memory sample analysis: added pivot_result and filter_result_rows on result_id, backed by the :shared Explore pivot and worksheet engines, plus export_result to write owner-only CSV exports under {dataDir}/exports/.
  • Workspace tools: added list_saved_queries, run_saved_query, save_query, delete_saved_query (two-step confirmation), list_recipes, apply_recipe, and list_query_history, all backed by the shared specToSql compiler.
  • MCP resources and prompts: added safedb://connections, safedb://connections/{connection_id}/tables, and safedb://results/{result_id}/summary resources plus explore-database and analyze-result prompts.
  • CLI enhancements: added safe-db-mcp connections test [<id>] and safe-db-mcp settings (show, risk-gate, default-connection, clear-default-connection, block-schema, unblock-schema); added safe-db-mcp recipes list, recipes import <absolute path>, and recipes export <recipe_id>.
  • Case-insensitive catalog lookup for describe_table and find_join_paths, with up to 5 suggested tables when a name is unknown and invalid_arguments listing both names when a reference is ambiguous.
  • Db2 internal catalog tables (EXPLAIN_*, ADVISE_*) are hidden from list_tables, search_catalog, describe_table, find_join_paths, and the tables resource.
  • find_join_paths accepts foreign_key_only to exclude indexed-column candidates.
  • Unknown-column errors on filter_result_rows, pivot_result, and sorts list known result columns.
  • run_query, run_saved_query, and assess_query receipts include dialect; capped samples report sample_cap_reason (sql_limit or byte_budget) and unbounded SQL sets sample_unbounded.
  • Additive structuredContent for object tools and published outputSchema for all 16 object tools: list_tables, describe_table, search_catalog, find_join_paths, run_query, assess_query, get_result_rows, summarize_result, pivot_result, filter_result_rows, apply_recipe, run_saved_query, save_query, delete_connection, delete_saved_query, and export_result.
  • Intentional MCP contract changes: a visualization-only apply_recipe (no pivot and no worksheet) is a validation error; worksheet receipt column names are source columns, not display labels; analyze-result result_id is optional.
  • Unknown-table suggestions rank visible tables by spelling similarity (edit distance / prefix) instead of alphabetical order.
  • run_query / assess_query risk reasons include the same scoring evidence on allowed queries as on blocked queries.
  • Empty run_query samples keep JDBC column types (instead of "unknown") and omit sample_unbounded plus the large-limit warning when the result is known empty.
  • Pivot dimension names use table-aware display labels for default, explicit, and recipe pivots (strip {alias}__; qualify collisions as table.column).
  • Pivot measure names use the same table-aware display labels (Sum amount, Max orders.id on collision).
  • Multi-dimension pivots emit one output column per row dimension; totals stay in the totaled dimension instead of collapsing into a single header.
  • filter_result_rows members / included_keys matches actual cell values (view, 1) as well as internal TextCell:view keys.
  • Failure to parse a MySQL JSON plan, including Impossible WHERE / no matching const-table plans, no longer forces confirmation_required on ordinary gated SELECTs.
  • setup / connections add refuse to save a connection when SAFEDB_KEYCHAIN_BACKEND=disabled; connections test then says to unset that override rather than add the connection again.
  • Empty run_query samples drop the leftover material-payload risk score/reason.
  • Duplicate “joins an additional relation” risk reasons collapse to one signal.
  • Duplicate “high-row scan of a large table” risk reasons from a self-join collapse to one reason; both sides still count toward the score.
  • Multi-dimension pivots no longer emit empty parent-group header rows with a null child and null measures, including when show_subtotals is false.
  • A mismatched destructive confirmation no longer replaces a still-valid outstanding token.
  • filter_result_rows documents that members takes included_keys, not value.
  • runQueryCore fails closed with QueryError.RiskGate (risk_gate) when risk assessment is still pending (AssessmentPending).
  • LIKE / ILIKE / NOT LIKE ... ESCAPE '!' round-trips through SQL parsing, FilterSpec, and specToSql / saved queries (and clears likeEscape when changing columns or operators in the Query Builder).

safe-db v1.0.1

Choose a tag to compare

@COBOL-Kid COBOL-Kid released this 08 Nov 20:45
5aba6fb

Changelog

  • History: Builder and SQL runs appear as they complete, without restarting the app. Delete a single entry, or clear the whole list.
  • Light mode default: New installs start in light mode. A missing or unrecognized theme value also starts light. An explicitly saved dark theme stays dark.
  • Query builder layout: The toolbar wraps the connection controls and Run as groups at the 960x600 minimum window with the sidebar expanded. Joins stay reachable at that size, and Query options opens a dialog with distinct rows, grouping, sorting, and joins. Filters sit in a height-bounded scrolling strip above the canvas, so a long IN list cannot cover the table cards.
  • Filters: Filter chips stay visible at the default width with the sidebar expanded. BETWEEN fields stay together with their from, and, and to labels, and the operator menu no longer truncates BETWEEN. Each wrapped IN value keeps its remove button, and timestamp values stay readable.
  • Table cards: Column names show at every width. A card shows column types once it is wide enough, and hovering a column name reveals the type on a narrower card. Each column toggle has an accessible label.
  • SQL workspace:. The header wraps the title, connection and schema pickers, and Run as groups at the 960x600 minimum. With no saved connection selected, the screen shows only the Choose a database to query prompt and an Open Connections action.

safe-db v1.0.0

Choose a tag to compare

@COBOL-Kid COBOL-Kid released this 06 Nov 17:41
687b8f8

Safe DB 1.0.0 is the first official release: a desktop app and MCP server (safe-db-mcp / @safe-db/mcp) that give people and AI agents the same bounded, read-only access to databases.

Queries run through a shared engine with connection timeouts, result-size limits, and a risk gate that can stop high-risk plans before they execute. Failures come back as structured errors instead of an unbounded session.

Downloads, documentation, exact query limits, and supported databases are at https://safe-db.dev