opensolr-mcp

Index plain-text documents via the Opensolr Data Ingestion API. Ingestion is ASYNC: embeddings, sentiment, and all derived fields are computed server-side and documents become searchable within about a minute (progress is visible in the Opensolr Control Panel). With wait=true (default) this blocks until the job completes. Metadata keys become filterable meta_* fields; metadata "uri" (a real URL) is used as the document identity — the Solr id is md5(uri). Returns job info and the resulting Solr document ids.

Ask a question and get a grounded RAG answer generated ONLY from the content already indexed in the given Opensolr index. Retrieval runs through the platform's tuned hybrid pipeline (the index's saved Search Tuning applies automatically) — the top rag_docs hits (first rag_words words of text each) become the LLM context, the same pipeline as the hosted search UI. filter_query optionally narrows retrieval with a raw Solr fq expression; instruction optionally replaces the default prompt (e.g. "Answer in German", "Extract a list of people"); tuning optionally overrides retrieval knobs per call. That list is the whole set, not a sample — an abbreviated one reads as everything that is supported, and freshness_boost was invisible to callers because of it: fw_title, fw_description, fw_uri, fw_text, fw_text_t, lexical_weight, vector_weight, vector_topk, search_mode (union / keywords_required / meaning_required / intersection), quality_boost, min_score, freshness_boost, fresh_bias, lexical_norm_k, mm (flexible / balanced / strict or raw Solr mm syntax). freshness_boost and fresh_bias are different knobs despite the names: the first is a hard window in DAYS that filters older documents out, the second only re-orders, multiplying each score by a recency curve on creation_date so recent documents win ties while nothing becomes unreachable.

Install the Opensolr vector schema on an index of a vector region, so Data Ingestion, vector and hybrid search and PDF pages all work on it. It replaces the index configuration. An index that already holds documents needs force=true: those documents keep the fields they were written with and must be sent again. An index outside the vector regions cannot take it (the answer says where to create one). Returns the new check.

Whether an index is ready for every Opensolr tool, and exactly why not. Answers "ready", "capabilities" (ingestion, vector_search, pdf_pages, keyword_search), the server and its Solr version, whether the server is in a vector region, the fields and types that are missing or different from the Opensolr vector schema, "problems" (sentences) and "fix": "apply_schema" (run opensolr_apply_schema), "new_index" (create one with opensolr_create_index in a region of opensolr_vector_regions) or "none". The other tools run this check once per index on their own; call it to see the details.

Create a new vector-enabled Opensolr index in a vector region. location: an environment id from opensolr_vector_regions, or a short name matched against that live list by environment id or country (e.g. "us", "de"). Additional dedicated regions can be deployed on request ([email protected]).

Delete documents by ids (Solr ids or your original ids) or by a raw Solr query, e.g. 'meta_category:"drafts"' or '+id:"abc123"'.

What a picture shows and what is written on it: "caption" (a sentence about what it depicts), "text" (the text read off it by OCR, empty if none), "labels" (the things in it) and "codes" (barcodes / QR codes). index is the Opensolr index the reading is counted on. Nothing is stored.

Read local files on Opensolr without indexing them: their text, the text in their pictures (OCR, 100+ languages), a PDF page by page, scanned pages and picture files (JPG, PNG, TIFF...) included. path is a file or a folder (pattern filters the files of a folder, e.g. "*.pdf"; recursive walks its sub-folders). index is the Opensolr index the work is counted on (pictures read: 0.5 AI request each; a file read before is free). Files still being read after wait_seconds come back with state "queued"/"processing": call again with the same path to get them — they are not sent twice. Each file: path, state, mime and its "pages" (a PDF: page, text, is_ocr, lang) or "text"; "error" when it could not be read. The whole answer is cut at max_chars characters.

Index local files into an Opensolr index: one file or a whole folder. Each file is read on Opensolr — its text, the text in its pictures (OCR), the pictures without text described — and a PDF is indexed ONE DOCUMENT PER PAGE, so a search finds the page that answers it (results carry the page number and a link to it). Word, RTF, OpenDocument, text, HTML and picture files (scans) are one document each. pattern filters the files of a folder (e.g. "*.pdf"), recursive walks its sub-folders, metadata is added to every document as filterable meta_* fields. Sending the same files again replaces them. Files still being read after wait_seconds come back under "pending": call again with the same path to index them (the bytes are not sent twice). Returns "documents" (path, source, uri, id), "pending", "failed" and "jobs" (Data Ingestion job ids, see opensolr_ingest_status); documents become searchable about a minute after their job runs.

Get connection details for an index: Solr URL, version, environment. (Credentials are intentionally not returned.)

Status of a Data Ingestion job (state, processed/success/failed doc counts). Also visible in the Opensolr Control Panel.

List all search indexes in the connected Opensolr account.

Read one indexed document in full — search results carry only an excerpt. document is the "document_id" or "id" of a search result, the id it was sent with, or its URL. A PDF indexed page by page comes back as its pages in order (page_from to page_to, default the first 50; "pages_indexed" tells how many it has) — read the next ones with page_from. Any other document comes back as one text. The answer is cut at max_chars characters ("truncated": true).

Search an Opensolr index and return the k most relevant documents. search_mode: "hybrid" (default — BM25 + semantic kNN fused per document), "semantic" (pure kNN), or "lexical" (pure keyword edismax — no embedding call, zero AI quota, works on ANY index including non-vector ones). For hybrid: mode is union / keywords_required / meaning_required / intersection, alpha balances semantic (0) vs lexical (1). filter_query accepts a raw Solr fq expression, e.g. 'meta_category:"docs"'. fresh_bias biases the ranking toward recent documents by multiplying each score by a recency curve on creation_date. It re-orders and never filters: the hit count is unchanged, nothing old becomes unreachable, and a document with no creation_date is simply left unboosted. Works in all three search modes. Off by default — turn it on when newer should beat older on a tie. date_from / date_to (YYYY-MM-DD, both included, either may be left out) keep only documents dated in that period (creation_date) — use them when the question names a period. A PDF indexed page by page is one result per PDF (group_pages, on by default): its best page, with "uri", "page", "page_url" (opens the PDF at that page), "document_id" (for opensolr_read_document) and "more_pages", its other matching pages.

Search an Opensolr index with a PHOTO instead of a text query. The image at image_path is read three ways by the Opensolr image engine — visual labels (what it depicts), OCR text (words printed on it), and any barcode / QR code — and turned into a text query that runs through the normal search. No image vector is stored; the picture simply becomes words. using selects which reading drives the search: "auto" the engine's chosen text (OCR text when the picture is mostly text, otherwise the visual labels) — the default, "meaning" the visual labels (what the picture depicts), "text" only the OCR text read off the picture (empty if none), "code" the first barcode / QR code, matched as an exact keyword. search_mode / mode / alpha / fresh_bias / filter_query / date_from / date_to / group_pages behave exactly as in opensolr_search. Returns {"read": {text, mode, labels, codes}, "results": [...]} so the caller sees both what the picture was read as and the matching documents.

List the vector-enabled Opensolr environments currently available (the servers with the Opensolr web crawler that are up, any Solr version: dense vectors, the hybrid query parser and Data Ingestion).