LambdaDB MCP 0.1.4

Changes

• Standardize feature terminology as native embedding and native reranking across MCP guidance, documentation and live-test descriptions.
• Pin LambdaDB TypeScript SDK 0.8.1 and MCP SDK 1.31.0. The MCP SDK update includes the fix for GHSA-6qxp-vccf-f47h; this stdio server does not use the affected HTTP OAuth client path.
• Include the reviewed proxy-addr 2.0.8 lockfile update.
• Preserve public tool identifiers, wire fields such as managedEmbedding, configuration keys, legacy application errors, read-only defaults and pinned source links.

Compatibility

The MCP SDK now limits the stdio receive buffer to 10 MiB and closes the transport on overflow. SDK-generated schema-validation errors include field paths. Existing MCP tool schemas and application defaults remain unchanged. Node.js >=22.14.0 is required.

Validation

The stable candidate passed version/type/build checks, 56 local tests and 35 installed-package MCP/stdio tests. npm audit reported zero vulnerabilities. PR and main CI passed Node.js 22.14.0, 22 and 24; automatic review completed without findings. No new paid or state-changing live smoke was run. The release workflow completed successfully. The published npm tarball passed 35 clean-consumer MCP/stdio tests and a zero-vulnerability audit; its integrity, source commit, SDK pins and provenance statement metadata match the release.

Install

npx --yes @functional-systems/lambdadb-mcp@0.1.4

See #21 for candidate validation and review.

Published package: https://www.npmjs.com/package/@functional-systems/lambdadb-mcp/v/0.1.4

Publication workflow: https://github.com/lambdadb/lambdadb-mcp/actions/runs/37600343928

LambdaDB MCP 0.1.3

Changes

• Pin the LambdaDB TypeScript SDK to 0.8.0 and support Bayesian hybrid search through the existing MCP query tool.
• Forward top-level candidateSize for Bayesian retrieval without rerank. With rerank, use rerank.candidateSize; existing defaults, free-form queries and server error handling are preserved.
• Support native vector embedding configuration without managedEmbedding, retaining explicit managedEmbedding: true for older servers and SDK rejection of contradictory settings.
• Preserve rerank status metadata and retrievalScore, including SDK-managed docsUrl downloads. Existing tool names, read-only defaults and write opt-in are unchanged.
• Add practical MCP examples, focused contract tests and an opt-in Bayesian/native live suite with verified temporary-data cleanup.

Compatibility and validation

Bayesian/native embedding additions target backend contract 9072a1bc8925954369a887f558f1eaf387b7ea0e and require a supporting deployment. Older servers require the explicit legacy embedding flag. Existing collections and indexes are not migrated.

The reviewed implementation passed 56 local tests, 35 clean-consumer stdio tests and three authorized development-service suites covering real embeddings, Bayesian budgets, managed reranking, downloads and cleanup. Stable candidate metadata and its retained tarball passed release/package checks; the release workflow validates and publishes its own exact tested artifact. These checks do not establish production deployment, search quality or load behavior.

• Release PR #15
• MCP examples
• Validation evidence

npx --yes @functional-systems/lambdadb-mcp@0.1.3

LambdaDB MCP 0.1.2

Pin the LambdaDB TypeScript SDK to @functional-systems/lambdadb@0.7.0 and add optional per-query managed reranking to lambdadb_query_collection.

• Support all 49 fixed lowercase text analyzer presets while preserving omitted standard defaults and analyzer serialization. The keyword text analyzer remains distinct from the keyword field type.
• Expose the fixed typesafe / jev-1.13.0 model, query text, scalar text fields, candidate cap, failure policy and optional custom criteria. Keep final size, vector k and candidateSize distinct; provider credentials are managed by LambdaDB.
• Preserve envelope evaluation/retrieval scores, numeric zero, precision, server order and rerank status metadata through inline and docsUrl responses. Omitted/null rerank preserves existing searches; existing sort, query, facet and ref restrictions remain.
• Add a separately opted-in live reranking suite with owned-resource cleanup. Read-only defaults and explicit write opt-in remain unchanged.

Validation: 52 local tests, 31 retained-candidate consumer/stdio tests, and Node 22.14.0 / 22 / 24 CI passed. Both authorized development-service live suites passed, including actual default/custom inference and docsUrl envelope preservation; temporary collections were deleted and absence verified. See candidate and published-dev validation.

Features require supporting backend deployment/model availability. These checks do not establish production deployment, search quality, load/failure coverage or billing readiness. Existing collections/indexes are not migrated, and existing MCP installations require an update to this MCP release. This server uses the native SDK and does not add a Qdrant adapter.

The release workflow validates and publishes its exact tested artifact to the npm latest channel. Workflow and registry verification are separate from the candidate results above.

LambdaDB MCP 0.1.1

LambdaDB MCP 0.1.1 adopts the official LambdaDB TypeScript SDK 0.6.0 and adds keyword facets and match-all queries.

Added

• lambdadb_query_collection accepts keyword facets, omitted query for match-all, and size: 0 for facet-only requests.
• Up to five facet fields, including dotted paths. Bucket size accepts 1–100, omitted, or null; omitted/null uses the server default of 10.
• Facet results are preserved through inline responses and automatic docsUrl downloads.
• All 16 text analyzers are accepted by collection creation, including the 12 new analyzers in SDK 0.6.0.

Existing search options, ref/consistent-read semantics, read-only defaults, and explicit write opt-in are preserved. MCP JSON Schema and actual tool-to-SDK request/response paths are covered by regression tests.

Compatibility

Facets require a supporting server deployment and newly built keyword indexes. This release does not migrate existing Collections, indexes, or Tags. Update the MCP package itself to receive these changes; publishing the SDK alone does not update existing MCP installations.

Requires Node.js >=22.14.0.

npx --yes @functional-systems/lambdadb-mcp@0.1.1

Validation

• Local checks: 48 tests passed.
• Exact candidate tarball: 27 installed stdio tests passed.
• Node 22.14.0 / 22 / 24 CI passed for the release PR and main candidate.
• Development-service live tests passed, including facets, analyzers, document downloads, refs, and temporary Collection deletion/absence verification.

The GitHub Release workflow publishes to npm using OIDC and verifies the exact publication artifact. See the candidate validation record for validation scopes.

Full changelog: v0.1.0...v0.1.1

LambdaDB MCP 0.1.0

LambdaDB MCP 0.1.0

First stable npm release of the LambdaDB MCP server.

• Run @functional-systems/lambdadb-mcp through the lambdadb-mcp executable without cloning or building the repository.
• Requires Node.js >=22.14.0; validation covers the exact minimum and current Node 22/24 LTS.
• Preserves project-scoped environment variables, read-only defaults and explicit write opt-in.
• Reports the installed package version and shuts down cleanly on stdio EOF or termination.
• Adds dev/rc/stable release automation with npm Trusted Publishing/OIDC and exact-tarball consumer checks.

After publication succeeds, configure the connection environment variables and run:

npx --yes @functional-systems/lambdadb-mcp@0.1.0

The stable release publishes to latest; unqualified npx resolves the stable package after registry verification.

Candidate validation passed 44 source tests, 23 installed stdio tests, and an actual stdio smoke against an explicitly designated development project. Temporary collection cleanup was verified. The release workflow must validate its final publication artifact separately.