Conversation
…ring Adds an interchangeable MCP transport for OKP RAG retrieval alongside the existing OGX/Solr vector_io path, forked at build_rag_context by okp_rag_mcp_enabled(). - OkpMcpConfiguration (rag.okp.mcp): enabled flag, url, tool_name, max_chunks, timeout, authorization_headers, plus structured product/product_version filters that mirror the Solr transport's chunk_filter_query filtering. - OkpMcpRetriever + call_okp_search: single direct MCP search-tool call over streamable HTTP (pydantic-ai MCPToolset), mapping results to the backend-neutral RAGChunk/ReferencedDocument contract shared with the Solr path; product/product_version passed as structured tool args when set. - Regenerated OpenAPI schema for the new config fields. - Unit tests for the client, provider, config model, configuration fork, vector_search fork, and enrichment. The server-side product/version filtering support (mimcp search tool) is a separate change in the RHOKP repository. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Introduce OkpFilter/OkpProductFilter, a backend-neutral query-time RAG
filter (product selections, each scoping its own exact-match versions),
decoupled from Solr fq and the OGX {type,key,value} grammar so the public
interface survives the OGX->pydantic-ai migration.
- models.common.query: OkpProductFilter + OkpFilter (nested array shape).
- QueryRequest / OpenAI responses request: new optional `okp` field.
- OkpMcpRetriever.fetch: accept `okp`, fan out one MCP search per
(product, version) pair, then merge/dedup/sort/cap results.
WIP: `okp` is not yet threaded through vector_search.build_rag_context ->
_fetch_okp_rag_mcp/_fetch_okp_rag, so the field is accepted but not yet
consumed. Endpoint wiring, Solr-side translation, tests, and rag_guide
docs still pending.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Thread the request-level `okp` filter from all three inference entry points (query, streaming_query, responses) through build_rag_context to whichever OKP transport is active, completing the query-time filtering started in the previous commit. - vector_search: build_rag_context/_fetch_okp_rag_mcp/_fetch_okp_rag now accept `okp`; _okp_filter_to_structured translates it to an OGX eq/in/and/or filter for the legacy Solr path (AND-combined with any structured solr filter), so `okp` is not a no-op on the default transport. - MCP transport: OkpMcpRetriever.fetch fans out one search per (product, version), then merges/dedups/score-sorts/caps; a query-time filter overrides the launch-time config defaults. - Regenerate OpenAPI schema (OkpFilter/OkpProductFilter components) and document `okp` in rag_guide.md as the backend-neutral, preferred filter. - Unit tests: model validation, Solr translation, param merge, provider fan-out/override/merge/partial-failure, and transport forwarding. Verified: ruff, black, pydocstyle, pylint, pyright, and mypy clean on changed sources; affected unit suites pass. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ansport probe
Replace the launch-time MCP config flag with automatic per-request transport
selection. The OKP config reverts to its pre-MCP shape (rag.okp.{rhokp_url,
offline, ...}) with no mcp block; the MCP endpoint is always derived as
rhokp_url/mcp. Old configs keep working unchanged.
Transport selection:
- Launch: always wire the Solr vector_io provider (no synthesis-time coupling
to RHOKP; enrich_okp_mcp and the synthesis fork are removed).
- Query: prefer the MCP transport whenever okp_mcp_available() is True, falling
back to Solr when MCP is unavailable or hard-fails for the request.
- Fail-forward: the probe result is TTL-cached (not sticky) so an upgraded
RHOKP is adopted without restarting LCORE.
Product/version filtering is query-time only via the request `okp` filter; the
launch-time product/product_version config is dropped. OkpMcpConfiguration and
RH_SERVER_OKP_MCP_DEFAULT_URL are removed; OpenAPI schema regenerated.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Contributor
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: trueThanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
The RHOKP MCP `search` tool now accepts a structured, Solr-fq-analogous product filter and builds the query-side filter itself, so the OKP MCP retriever no longer fans out one call per (product, version) combo. - _client.call_okp_search: replace scalar product/product_version with an optional structured `products` arg, forwarded verbatim and omitted when None/empty. - _provider: drop _resolve_search_combos; add _okp_products_arg to translate the transport-neutral OkpFilter into the structured products list, and issue a single search call, raising OkpMcpUnavailableError on any failure so the caller falls back to the Solr transport. - Rewrite unit tests for the single-call structured-filter contract. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
mwcz
commented
Sep 17, 2026
| "description": "Exact product identifier (exact match, no wildcards).", | ||
| "examples": [ | ||
| "openshift_container_platform", | ||
| "rhel" |
The OKP product identifier is red_hat_enterprise_linux, not rhel. Correct the OkpProductFilter/OkpFilter examples and the tests exercising the slug, and regenerate the affected openapi.json examples from the model. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
mwcz
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
RHOKP RAG context used to enrich LCORE responses currently comes from direct calls to the Solr API hosted in RHOKP.
This PR adds conditional support for fetching that same context via RHOKP's upcoming MCP server. The support is conditional in the following sense: when LCORE is configured to fetch context from RHOKP, it will attempt MCP, and fall back to Solr if MCP is not available. In this way, LCORE has support for RHOKP RAG regardless of the version of RHOKP currently deployed. The conditional support also allows existing LCORE configuration to work with RHOKP before and after the release of the MCP server.
The plan discussed with the LCORE team is that the Solr provider (in lightspeed-providers) will be deprecated and eventually removed. The migration to Pydantic AI will not include Solr, so the migration timeline is, in a way, the same as the timeline for Solr provider removal.
Transport selection
vector_ioprovider.okp_mcp_available()isTrue, falling back to Solr when MCP is unavailable or hard-fails for the request. This provides compatibility with RHOKP RAGProduct/version filtering is now query-time only via the request
okpfilter; the launch-timeproduct/product_versionconfig is dropped.OkpMcpConfigurationandRH_SERVER_OKP_MCP_DEFAULT_URLare removed; the OpenAPI schema is regenerated.Testing
uv run make format,black,pylint(10.00/10),pyright src(0 errors): clean.uv run make test-unit: 3600 passed, 1 skipped.make verifystill fails oncheck-types-tests(mypy-on-tests): 41 pre-existing errors across 13 files not touched here (identical at the branch base).The RHOKP team can provide a container image of RHOKP with RAG and MCP built in (but with a small sample of the full corpus to save on build & transfer time), and we will also work on updating the internal instance of RHOKP to include MCP as well, for even easier testing.
🤖 Assisted by Claude Code