What googleapis/mcp-toolbox shipped
Written by FoxPlug from the repository's own releases; not affiliated with Mcp-toolbox. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/googleapis/mcp-toolbox; Mcp-toolbox did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- Released version 1.16.0, which adds skills over MCP extension support, a new mssql function for returning source code for custom database objects, and removal of the dataplex
search_aspect_typestool from the prebuilt config. Release - Fixed tools.yaml reload watching so that changes to the default config file are detected and reloaded even when no config flag is passed. Pull request #4025
- Fixed the server/discover result to include ttlMs and cacheScope, which the MCP 2026-07-28 spec requires on DiscoverResult. Pull request #4428
- Fixed HTTP header name handling so that names differing only in casing are canonicalized at each merge layer, preventing a lower-priority value from overriding an intended header. Pull request #4417
- Added the mssql-get-database-object-source-code tool for Microsoft SQL Server and Cloud SQL SQL Server to retrieve SQL definitions of custom database objects. Pull request #3954
- Fixed the Oracle integration step in CI so that it fails when go test fails, instead of only computing coverage. Pull request #3997
- Annotated text/template imports with NOLINT comments in the Cloud SQL Admin, HTTP and parameters files. Pull request #4425
- Updated Hugo for release v1.16. Pull request #4426
Why it matters
This week brings a new release with a skills extension and a new tool for reading custom object definitions from SQL Server, alongside fixes for config reloading and header precedence that affect everyday runs. The MCP discover result now carries the required caching fields, and integration tests for several sources moved to the MCP endpoint. CI also fails correctly when Oracle tests fail, so broken builds are caught instead of passing silently.
Changelog entry
- Release v1.16.0: adds skills over MCP extension, new mssql source code function, removes dataplex
search_aspect_typesfrom prebuilt config. Release - Watch tools.yaml for reload when the config flag is omitted. Pull request #4025
- Include ttlMs and cacheScope in the server/discover result. Pull request #4428
- Canonicalize HTTP header names per merge layer to preserve case-insensitive precedence. Pull request #4417
- Add mssql-get-database-object-source-code tool for mssql and cloudsqlmssql. Pull request #3954
- Fail the Oracle integration CI step when go test fails. Pull request #3997
- Add NOLINT comments to non-YAML text/template imports. Pull request #4425
- Update hugo for release v1.16. Pull request #4426
- Automate public-image and deprecation tagging in artifact promotion. Pull request #4324
- Migrate TiDB integration tests to testcontainers. Pull request #4360
- Add CockroachDB MCP integration tests. Pull request #4113
- Add TiDB MCP integration tests. Pull request #4095
V1.16.0 is out: skills over MCP extension, a new mssql tool for custom object source code, and dataplex prebuilt config cleanup. Also fixed this week: tools.yaml reload without the config flag, header name casing precedence, and the server/discover caching fields.
Version 1.16.0 shipped this week with support for the skills over MCP extension, a new mssql tool that returns source code for custom database objects, and removal of the search_aspect_types tool from the dataplex prebuilt config. Fixes include tools.yaml now reloading when the config flag is omitted, HTTP header names with different casing no longer overriding each other during merges, and the server/discover result including the ttlMs and cacheScope fields required by the current spec. Integration tests for several sources moved to the MCP endpoint, TiDB tests now use testcontainers, and the Oracle CI step now fails when go test fails rather than reporting coverage alone.
Week of October 26, 2026
What shipped
- Added dynamic UI resources with a NewDynamicUIResource constructor and a resources.ResourceProvider interface, letting built-in tools register ui:// resources that fetch or generate HTML at resources/read time, plus thread-safe CSP get and set. Pull request #4315
- Made writeMode protected in bigquery-execute-sql fail closed for non-SELECT statements, validating session metadata before any dry-run or query. Pull request #4359
- Set UseLegacySql to false when creating a BigQuery session for writeMode protected sources so session creation matches subsequent queries. Pull request #4358
- Fixed BigQuery sources to recover default application default credential clients after on-disk re-authentication, carrying over the original contributor implementation. Pull request #4329
- Compacted zero-valued optional schema fields in bigquery-get-table-info output while always keeping Name and Type and preserving top-level metadata and key ordering. Pull request #4328
- Extended the Looker dashboard element tools to create, update and read text tiles covering plain text, Looker-flavored Markdown and HTML, and Slate rich-text headers. Pull request #4180
- Added debug logging across the BigQuery tools, including sql, execute-sql, forecast, analyze-contribution and the dataset and table info and listing tools. Pull request #4346
- Trimmed the secure content scan allowed prefixes in CI to the release sources the pipeline actually pulls from. Pull request #4339
Why it matters
This week tightened BigQuery safety around protected write mode and session creation while making bigquery-get-table-info output cleaner. It also broadened what can be built and shown: dynamic UI resources for built-in tools and text, markdown and rich-text Looker dashboard tiles. BigQuery debugging and credential recovery after re-auth also improved.
Changelog entry
- Add DynamicUIResource and ResourceProvider so tools can register ui:// resources that generate HTML at read time, with thread-safe CSP access. Pull request #4315
- Fail closed in bigquery-execute-sql writeMode protected for non-SELECT statements, validating session metadata before running queries. Pull request #4359
- Set UseLegacySql to false when creating BigQuery sessions for protected write mode sources. Pull request #4358
- Recover default ADC clients after on-disk re-authentication for BigQuery sources. Pull request #4329
- Omit zero-valued optional schema fields in bigquery-get-table-info output while keeping Name and Type. Pull request #4328
- Support text, markdown and rich-text dashboard elements in the Looker add, update and get dashboard element tools. Pull request #4180
- Add debug logging across the BigQuery tools. Pull request #4346
- Trim secure content scan allowed prefixes in CI to release sources. Pull request #4339
This week: dynamic UI resources with a ResourceProvider interface, fail-closed writeMode protected for non-SELECT BigQuery statements, LegacySQL fixed on session creation, ADC client recovery after re-auth, compacted table-info schema output, and text and rich-text Looker dashboa
This week in the project: built-in tools can now register dynamic ui:// resources through NewDynamicUIResource and the resources.ResourceProvider interface, with thread-safe CSP handling. BigQuery writeMode protected now fails closed for non-SELECT statements, session creation sets UseLegacySql to false, and default ADC clients are recovered after on-disk re-auth. bigquery-get-table-info omits zero-valued optional schema fields while keeping Name and Type. Looker dashboard element tools now handle plain text, markdown and rich-text tiles. Debug logging spans the BigQuery tools, and CI scan prefixes were trimmed to release sources.
Week of October 19, 2026
What shipped
- Version 1.15.0 was released, adding data agent create, update and delete tools for conversational analytics, saved query execution by
query_idin the Looker query tool, and backported MCP Apps UI support. Release - MCP Apps UI support was backported across protocol versions v20241105 through v20251125, so clients that negotiate earlier protocol versions can also use interactive UI resources and tool metadata. Pull request #4057
- The Toolbox playground UI now supports MCP Apps, detecting tools with UI resources, showing an APP badge, and mounting the interface in a sandboxed iframe. Pull request #4063
- An extension for serving skills over MCP was added, covering skill URI resources, discovery and validation. Pull request #4104
- The Looker query tool can now run saved queries by numeric ID or slug, with optional result format,
vis_configand drill link parameters. Pull request #4220 - A new looker-import-dashboard-from-lookml tool exposes the Looker ImportDashboardFromLookml endpoint in the Looker prebuilt toolset. Pull request #4156
- The
add_dashboard_elementandupdate_dashboard_elementactions now bind all dashboard filters in a single listener group instead of one entry per filter. Pull request #4326 - Authentication now checks that the issuer claim in a JWT matches the configured OIDC discovery issuer. Pull request #4296
- The web UI gained a collapsible sidebar and a resizable, scrollable tool description that preserves formatting for long descriptions. Pull request #4281
- Read failures on file resources no longer expose resolved server filesystem paths in error messages. Pull request #4260
Why it matters
This week brings a new release with new data agent tools and saved query execution for conversational analytics and Looker, plus MCP Apps UI support that now works for clients on older protocol versions. Skills can be served over MCP with documented layout and validation rules, and several fixes tighten authentication and error output. Users of Looker dashboards and the web UI also get fixes and layout improvements.
Changelog entry
- Release v1.15.0 with data agent tools, saved query execution by
query_idand MCP Apps UI backport. Release - Backport MCP Apps UI support across protocol versions v20241105 to v20251125. Pull request #4057
- Add MCP Apps support to the Toolbox playground UI with an APP badge and sandboxed iframe. Pull request #4063
- Add skills over MCP extension with skill URI resource support, discovery and validation. Pull request #4104
- Support running saved Looker queries by
query_idin the query tool. Pull request #4220 - Add looker-import-dashboard-from-lookml tool. Pull request #4156
- Bind all dashboard filters in
add_dashboard_elementandupdate_dashboard_element. Pull request #4326 - Validate the issuer claim in GetClaimsFromHeader against the configured OIDC issuer. Pull request #4296
- Add collapsible sidebar and resizable scrollable tool description to the web UI. Pull request #4281
- Sanitize server filesystem paths from file resource read errors. Pull request #4260
- Remove
search_aspect_typestool from the prebuilt dataplex config. Pull request #4193 - Let a skill publish the dynamic marker in place of a digest manifest. Pull request #4105
V1.15.0 is out: data agent create, update and delete tools, saved query execution by query_id in the Looker query tool, and MCP Apps UI support backported across protocol versions. Skills can now be served over MCP, and the playground UI renders MCP Apps.
V1.15.0 is released. It adds create, update and delete tools for data agents in conversational analytics, support for running saved Looker queries by query_id with optional result format and drill links, and MCP Apps UI support backported across protocol versions from v20241105 to v20251125, so clients on older protocols can use interactive UI resources too. Skills can now be served over MCP, with documentation covering the skill URI convention, layout and startup validation. The playground UI renders MCP Apps in a sandboxed iframe, and the tool inspector gained a collapsible sidebar and scrollable descriptions. Fixes this week bind all dashboard filters in dashboard element actions, validate the issuer claim in JWT handling, and keep server filesystem paths out of file resource read error
Week of October 12, 2026
What shipped
- The authorized method was removed from the tool interface so the authRequired check is enforced centrally instead of being silently accepted but not enforced on 25 tools. Pull request #4254
- Allowed datasets are now sorted when building BigQuery tool manifests, so repeated tools/list calls return a stable order. Pull request #4272
- Self-tracing is disabled on the GCP telemetry exporters, so enabling telemetry-gcp no longer has the exporter clients trace their own calls. Pull request #4230
- The CI pipeline now downloads zig from GCS to comply with MOSS. Pull request #4040
Why it matters
This week tightened how authentication is enforced across tools, closing a gap where a configured authRequired was shown in the manifest but not actually applied on many tools. It also added tools for managing data agents and made BigQuery tool manifests deterministic. Users of BigQuery sources and the conversational analytics source get more predictable listings and more management capability.
Changelog entry
- Remove authorized from the tool interface so the authRequired check is enforced in one place. Pull request #4254
- Sort BigQuery allowed datasets for deterministic tool manifests. Pull request #4272
- Disable self-tracing on GCP telemetry exporters. Pull request #4230
- Download zig from GCS in CI to comply with MOSS. Pull request #4040
Weekly update: auth enforcement is now centralized, so authRequired is actually enforced instead of only shown. Added create, update and delete tools for data agents. BigQuery manifests now list datasets in a stable order, and GCP telemetry exporters no longer trace themselves.
Week of October 5, 2026
What shipped
- The 1.14.0 release adds support for authorized views with dataset restrictions, includes LookML dashboards in the
get_dashboardstool, propagates the requester header on Firestore collection listing forget_schema, and carries other changes. Release - Skills list and skills get are now served over MCP with the skills extension declared on the 2026-07-28 protocol, grouping skills per request from the default group's resources so a reload shows up on the next request. Pull request #4050
- The looker
get_dashboardstool now includes LookML dashboards, which were previously excluded because the search only queried user-defined dashboards. Pull request #4216 - Config load now warns on skill resources that sit under no SKILL.md, listing orphaned URIs next to the duplicate-name warning. Pull request #4048
- Resource template URI schemes are now normalized to lowercase during validation, matching what resource URI validation already did. Pull request #4023
- Timeouts on both context and polling were increased for datalineage. Pull request #4227
- The cloud-gda integration test shard now also runs on conversational analytics changes and tolerates soft-deleted agents. Pull request #4160
- An Airlock setup script for Cloud Build was added, with APT, PIP and Docker container downloads proxied through Airlock and a.dockerignore to keep credentials out of the Docker context. Pull request #4144
- The Looker test expectation was updated because newer Looker instances return an empty list instead of an unsupported API error for legacy features. Pull request #4231
- Hugo was updated for the release. Pull request #4233
Why it matters
This week brings the 1.14.0 release, which adds authorized views with dataset restrictions, brings LookML dashboards into the get_dashboards tool, and fixes the requester header on Firestore collection listing. It also serves skills list and skills get over MCP and makes config load warn about skill resources that belong to no skill. Resource template URI schemes are now normalized so configs using mixed-case schemes validate as written.
Changelog entry
- Release: v1.14.0 with authorized views with dataset restrictions, LookML dashboards in
get_dashboards, and requester header propagation on Firestore collection listing forget_schemaRelease - Feat(skills): serve skills/list and skills/get over MCP with the skills extension on the 2026-07-28 protocol Pull request #4050
- Feat(looker): include LookML dashboards in
get_dashboardstool Pull request #4216 - Feat(skills): warn at config load on skill:// resources that belong to no skill Pull request #4048
- Fix(resources): normalize resource-template URI scheme to lowercase Pull request #4023
- Fix(datalineage): increase context and polling timeouts Pull request #4227
- Ci(cloudgda): run integration tests on conversational analytics changes and tolerate soft-deleted agents Pull request #4160
- Ci(moss): add artifact registry airlock proxies and a.dockerignore for the Docker context Pull request #4144
- Fix(looker): update looker test expectation for newer instances returning an empty list Pull request #4231
- Chore: update hugo for release Pull request #4233
V1.14.0 is out: authorized views with dataset restrictions, LookML dashboards in get_dashboards, requester header on Firestore collection listing for get_schema. Also serving skills/list and skills/get over MCP, lowercasing resource-template URI schemes, and warning on skill:// r
V1.14.0 shipped this week. Highlights: support for authorized views with dataset restrictions in BigQuery, LookML dashboards now included in the looker get_dashboards tool, and the requester header propagated on Firestore collection listing for get_schema. Beyond the release, skills/list and skills/get are now served over MCP on the 2026-07-28 protocol, config load warns about skill:// resources that belong to no skill, resource template URI schemes are normalized to lowercase during validation, and datalineage timeouts were increased. The cloud-gda integration shard now also runs on conversational analytics changes, and an Airlock setup script proxies package and container downloads in Cloud Build.
Week of September 28, 2026
What shipped
- Spanner source now supports Spanner Omni in addition to Cloud Spanner. Pull request #4198
- Looker tool adds looker-get-explore command to surface explore constraints. Pull request #4173
- Server validates all skills at startup to catch broken SKILL.md files and configuration issues. Pull request #4017
- SKILL.md files now publish under their frontmatter identity instead of generic resource name. Pull request #4044
- BigQuery tool checks dataset allowlist before dry run to return restriction errors instead of database errors. Pull request #4164
- Firestore source propagates requester header on collection listing for schema discovery. Pull request #4131
- Skills discovery now extracts and hashes skills from the resources map with frontmatter parsing. Pull request #4002
- BigQuery adds support for authorized views with dataset restrictions through dry run and reference auditing. Pull request #2561
- New docsite-link-sweep maintainer skill sweeps documentation for broken and non-canonical links. Pull request #4071
Why it matters
Skills are becoming first-class citizens with validation and discovery, while source and tool coverage expands to new services. Improved error handling in BigQuery and consistent behavior across data sources means fewer surprises for developers integrating these components.
Changelog entry
- Feat(sources/spanner): add support for Spanner Omni Pull request #4198
- Feat(skills): publish SKILL.md under its frontmatter identity Pull request #4044
- Fix(tool/bigquery): check lexically referenced datasets before dry run Pull request #4164
- Feat(tool/looker): add looker-get-explore tool Pull request #4173
- Feat(source/firestore): propagate requester header on collection listing for
get_schemaPull request #4131 - Feat(skills): validate skills at startup Pull request #4017
- Feat(skills): discover and hash skills from the resources map Pull request #4002
- Feat(bigquery): support authorized views with dataset restrictions Pull request #2561
- Fix(test/alloydbainl): match query value instead of model-chosen alias Pull request #4125
- Tests(bigtable): clean up all resources Pull request #4115
- Feat(skills): add SEP-2640 entry types, encoding, and validation Pull request #3991
This week: Spanner Omni support, Looker explore visibility, Bigtable/Firestore in data analytics, skill validation at startup, and skills publish under their own identity.
Week of September 21, 2026
What shipped
- Released v1.13.1 with a revert of the sources connection feature to fix issues in v1.13.0. Release
- Reverted the source connection on first use feature that was breaking oauth2 token sources. Pull request #4137
- Released v1.13.0 with support for deferring source connections until first use and a bigtable prebuilt config. Release
- File resources now accept skill:// URIs so agent skills can declare their resource files. Pull request #3989
- Looker tool now exposes
value_formatandvalue_format_namein field metadata for display formatting. Pull request #4094 - Conversational analytics tool now returns all data agents with support for manual paging instead of truncating results. Pull request #4066
- Fixed parseEnv to consistently substitute environment variables in YAML even when preceded by comment blocks. Pull request #4039
- Added.py to the allowed list of file type resources. Pull request #4121
- Added end-to-end evaluations for cloud-sql-postgres-observability prebuilt config covering all observability tools. Pull request #4020
- Removed non-deterministic assertions from Looker tests to improve test reliability. Pull request #4129
Why it matters
V1.13.1 addresses a critical regression in source connection handling that was breaking oauth2 sources. The fixes and improvements in the week enable better resource handling for agent skills, expanded tool capabilities for analytics and Looker, and broader test coverage for cloud-sql-postgres observability.
Changelog entry
- V1.13.1: Revert "feat(sources): connect sources on first use" to unbreak oauth2 token sources Release
- V1.13.0: Support --defer-source-connect flag for connecting sources on first use Release
- V1.13.0: Add prebuilt config for bigtable Release
- File resources now accept skill:// URIs for agent skill resource declarations Pull request #3989
- Looker tool exposes
value_formatandvalue_format_namein field metadata Pull request #4094 - Conversational analytics tool returns all data agents with optional
page_sizeandpage_tokenparameters Pull request #4066 - Fix parseEnv to consistently handle environment variable substitution with YAML comments Pull request #4039
- Add.py to allowed file type resource extensions Pull request #4121
- Add end-to-end evaluations for cloud-sql-postgres-observability covering
get_system_metricsandget_query_metricsPull request #4020 - Expand cloud-sql-postgres evaluations with health bloat and extension tuning scenarios Pull request #4022
V1.13.1 shipped to fix source connection issues. v1.13.0 added defer-connect, bigtable config, and expanded analytics/Looker tools. File resources now support skill:// URIs.
This week we shipped v1.13.1 to address source connection issues and v1.13.0 with new capabilities including deferred source connection, bigtable prebuilt config, and expanded tool features. File resources now accept skill:// URIs for agent skills, Looker tools expose additional field metadata, and analytics tools support pagination for complete data access. Environment variable parsing was also fixed for YAML comments.
Week of September 14, 2026
What shipped
- Added prebuilt configuration for Google Cloud Bigtable with 27 administration, schema discovery, and SQL tools. Pull request #4068
- Looker tools now expose
certification_metadatain dashboard and look payloads for enterprise content certification features. Pull request #4041 - Fixed looker-generate-embed-url tool to require type and id parameters instead of treating them as optional. Pull request #3913
- Clarified
execute_sqltool description in Cloud SQL PostgreSQL prebuilt config to explain dialect, return format, and usage guidance. Pull request #3884 - Fixed CockroachDB and Redis sources to properly release connection handles when connection attempts fail. Pull request #3933
- Added evals for cloud-sql-mssql covering table listing and SQL Server version queries. Pull request #3942
- Parameterized exit gate configuration to support testing release pipelines in sandbox environments. Pull request #4047
- Documented groups schema and specification for the com.google.cloud/toolbox.v1 extension. Pull request #3937
Why it matters
V1.12.0 expands integration support with new Bigtable configuration and improves existing integrations like Looker and Cloud SQL. Tool parameter handling and documentation refinements help prevent misuse and clarify capabilities for users.
Changelog entry
- Feat(bigtable): add prebuilt config for bigtable with 27 administration, schema discovery, and SQL tools Pull request #4068
- Feat(tool/looker): expose
certification_metadatain looker dashboard and look tools Pull request #4041 - Fix(tool/looker-generate-embed-url): require embed parameters instead of treating as optional Pull request #3913
- Fix(prebuilt/cloud-sql-postgres): clarify
execute_sqltool description Pull request #3884 - Fix(source/cockroachdb,source/redis): release the handle when a connect attempt fails Pull request #3933
- Ci(evals): add cloud-sql-mssql evals Pull request #3942
- Ci: parameterize exit gate configuration Pull request #4047
- Docs(extensions): add groups schema and specification for com.google.cloud/toolbox.v1 Pull request #3937