Fix search_repository: Docker URL, ecm:fulltext field, configurable highlights

[bdelbosc]

Summary

Three fixes and one improvement to search_repository and search_audit.

SUPINT-2477 — Fix search_repository/search_audit failing inside Docker

nuxeo.client.host stores the host as seen at connection time (http://localhost:8080), which does not resolve to the Nuxeo service from inside a Docker container.

Fix: use os.environ.get("NUXEO_URL", nuxeo.client.host) so the Docker service name can be injected via env var, with the client host as fallback. Also increases the ES probe timeout from 2s to 10s.

SUPINT-2478 — Fix fulltext search targeting non-existent ecm:fulltext field

ElasticsearchQueryBuilder was querying ecm:fulltext/ecm:fulltext.title and NaturalLanguageParser was requesting highlights on ecm:fulltext. These fields do not exist in the Nuxeo OpenSearch mapping — the catch-all field is all_field (a copy_to aggregate).

Fix:

• Query field: ecm:fulltext → all_field
• Highlight field: ecm:fulltext → ecm:binarytext with require_field_match: false (all_field has no store so cannot be highlighted directly)

SUPINT-2479 — Add configurable highlight fragment size and count

Fixed-size highlights (150 chars) were too small for meaningful content extraction from large PDFs. Added two optional parameters throughout the call chain (tools.py → es_passthrough.py → nl_parser.py):

• highlight_fragment_size (default: 150, max: ~9,000,000)
• highlight_number_of_fragments (default: 3; set to 0 for entire field)

Also adds source_fields passthrough for cases where raw _source content is needed.

Docs

• nuxeo_mcp_config.md: note on NUXEO_URL Docker service name requirement
• USAGE.md: examples for new search_repository parameters

[Copilot]

Pull request overview

This PR fixes Elasticsearch passthrough behavior for search_repository / search_audit in Dockerized deployments, aligns repository fulltext querying/highlighting with Nuxeo’s OpenSearch mapping, and adds configurable highlight/source-field controls to improve content extraction.

Changes:

• Use NUXEO_URL env var (fallback to nuxeo.client.host) for Elasticsearch passthrough base URL, and increase the repository ES probe timeout.
• Switch fulltext query target from ecm:fulltext to all_field, and update highlighting to target ecm:binarytext with require_field_match: false.
• Add source_fields, highlight_fragment_size, and highlight_number_of_fragments parameters through the tool → passthrough → parser stack; update docs and one unit test accordingly.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
USAGE.md	Documents new search_repository parameters for highlight sizing and _source field inclusion.
tests/test_es_query_builder.py	Updates expected fulltext query fields to all_field.
src/nuxeo_mcp/tools.py	Adds new search_repository parameters; uses NUXEO_URL for passthrough; increases repository ES probe timeout.
src/nuxeo_mcp/nl_parser.py	Updates highlight configuration (field + configurable fragment sizing).
src/nuxeo_mcp/es_query_builder.py	Changes default fulltext search field list to ["all_field"].
src/nuxeo_mcp/es_passthrough.py	Threads highlight params; adds _source passthrough and modifies result formatting.
nuxeo_mcp_config.md	Adds Docker note about setting NUXEO_URL to a reachable service hostname.
Dockerfile	Uses explicit octal permissions (--chmod=0755) for the entrypoint copy.

Comments suppressed due to low confidence (1)

src/nuxeo_mcp/es_passthrough.py:96

• source_fields is documented/used as “extra fields to include”, but it is passed straight into _source.includes. When a caller supplies only an extra field (e.g. ['ecm:binarytext']), Elasticsearch will omit the standard metadata fields (dc:title, ecm:path, etc.), and the formatted results will end up with empty strings for those base properties. Consider always including the required base fields plus any extras when building _source.includes (or treating source_fields as additive rather than replacing).

        # Parse natural language to Elasticsearch DSL
        es_request = self.nl_parser.parse_to_elasticsearch(
            query,
            index="repository",
            include_sort=True,
            include_pagination=True,
            include_highlight=True,
            highlight_fragment_size=highlight_fragment_size,
            highlight_number_of_fragments=highlight_number_of_fragments,
            apply_acl=True,
            user_principals=[principal] + groups,
            source_includes=source_fields,
        )

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.