Document Intelligence (PDF Search)

The Red Bricks MCP server includes four Document Intelligence tools that
search the full text extracted from project PDFs — brochures, features &
finishes sheets, incentive notices, price lists, deposit structures, and fact
sheets. Use them to answer questions that aren't in the structured project data:
appliance brand names, capped development-charge values, free-assignment terms,
finish details, and how incentives have evolved over time.

📘
These are MCP tools, invoked by an MCP client (e.g. Claude) connected to
the Red Bricks MCP server — not REST endpoints. See
Configure MCP Server for connection setup.

Availability

Plan	Document Intelligence
Sandbox / Essential	❌ Not available
Pro	✅ Searches the latest documents
Elite	✅ Latest + historical (older-revision) documents

On plans without access, the tools return a permission error. The
include_historical option only takes effect on Elite — on Pro it is silently
ignored (latest documents only).

How matching works

The PDF tools do exact keyword + synonym matching, not natural-language
search. Internally they run a two-stage match: a fast full-text pre-filter,
then a strict containment post-filter that drops phantom matches (and bridges
PDF-extraction encoding artifacts, so 30", 30-inch, and the mojibake 30�
all match 30 inch).

Practical implications for querying:

Open-ended queries ("luxury appliances", "premium finishes") won't match an
abstract noun. Expand them into concrete terms and call the tool a few times —
e.g. for "luxury appliances," search miele, sub-zero, wolf, thermador
separately and merge the results.
Sized queries ("30 inch range") — search the size synonym (30 inch); the
synonym groups cover hyphen/quote variants.
Phrase queries ("free assignment", "capped development charges") — the
synonym groups already cover common variants, so a single call is enough.

Coverage metadata

Every search response carries a meta.coverage block so you can tell "nothing
matched X" apart from "the index couldn't answer":

Field	Meaning
`total_active_documents`	Indexed docs in the searched scope
`documents_with_text`	Docs that extracted successfully and are searchable
`documents_with_no_text`	Docs that failed extraction or have no usable text
`pre_filter_candidates`	Rows the full-text pre-filter returned before strict filtering

If total_chunks_matched is 0 after a couple of query variants, treat it as a
genuine "not mentioned" — don't speculate.

`search-document-text`

Citation-backed chunk search within project PDFs (one project, or all projects).

Input

Field	Type	Required	Description
`query`	string	yes	Free-text phrase or keywords.
`project_id`	integer	no	Restrict to a single project.
`category`	string	no	Restrict to a category (see Categories).
`include_historical`	boolean	no	Include older-revision docs (Elite only).
`limit`	integer	no	Max chunks, `1`–`25` (default `10`).

Output

{
  "data": [
    {
      "chunk_id": 84213,
      "project_id": 6809,
      "project_name": "The Forest Hill Private Residences",
      "document_id": 1422,
      "document_name": "Features & Finishes.pdf",
      "category": "features-finishes",
      "page_number": 3,
      "excerpt": "...integrated Miele appliance package including a 30 inch gas cooktop...",
      "matched_term": "miele",
      "extraction_id": 5567,
      "stable_citation": {
        "document_id": 1422,
        "page_number": 3,
        "text_hash": "9f2c…"
      }
    }
  ],
  "meta": {
    "query": "miele",
    "expanded_terms": ["miele"],
    "total_chunks_matched": 4,
    "coverage": {
      "total_active_documents": 6210,
      "documents_with_text": 6151,
      "documents_with_no_text": 59,
      "pre_filter_candidates": 12
    }
  }
}

excerpt is a ~200-character window centered on the first match. Use chunk_id
with fetch-document-context to pull surrounding paragraphs.

`find-projects-mentioning`

Cross-project discovery — "which projects mention X?" Returns each matching
project with its match count and a sample excerpt.

Input

Field	Type	Required	Description
`query`	string	yes	Phrase or feature to search for across all projects.
`category`	string	no	Restrict to a category.
`include_historical`	boolean	no	Include older-revision docs (Elite only).
`limit`	integer	no	Max projects, `1`–`50` (default `25`).

Output

{
  "data": [
    {
      "project_id": 6809,
      "project_name": "The Forest Hill Private Residences",
      "match_count": 7,
      "categories_matched": ["brochure", "features-finishes"],
      "best_excerpt": "...purchasers receive free assignment and one year of free maintenance..."
    }
  ],
  "meta": {
    "query": "free assignment",
    "expanded_terms": ["free assignment", "assignment fee waived"],
    "total_projects": 12,
    "coverage": {
      "total_active_documents": 6210,
      "documents_with_text": 6151,
      "documents_with_no_text": 59,
      "pre_filter_candidates": 340
    }
  }
}

Projects are ranked by match_count (then relevance). total_projects is the
full count before the limit cap.

`list-project-documents`

Inventory of indexed PDFs for one project — what exists, their categories, and
extraction status. Useful before drilling into search-document-text.

Input

Field	Type	Required	Description
`project_id`	integer	yes	Project to list documents for.
`category`	string	no	Filter by category.
`include_historical`	boolean	no	Include older-revision docs (Elite only).

Output

{
  "data": [
    {
      "document_id": 1422,
      "name": "Features & Finishes.pdf",
      "category": "features-finishes",
      "type": null,
      "indexed": true,
      "extraction_status": "completed",
      "extraction_method": "pdftotext",
      "page_count": 4,
      "raw_text_length": 8231,
      "indexed_at": "2026-05-01T12:00:00+00:00"
    }
  ],
  "meta": {
    "project_id": 6809,
    "total_documents": 12,
    "include_historical": false
  }
}

Documents that failed extraction or are still pending are included with
indexed: false and their extraction_status, so you know why they might not
surface in search.

`fetch-document-context`

Expand a search hit by returning the chunks immediately before and after it, in
document order — for when an excerpt is suggestive but truncated.

Input

Field	Type	Required	Description
`chunk_id`	integer	yes	A `chunk_id` returned by `search-document-text`.
`before`	integer	no	Preceding chunks to include, `0`–`5` (default `1`).
`after`	integer	no	Following chunks to include, `0`–`5` (default `1`).

Output

{
  "data": [
    { "chunk_id": 84212, "chunk_index": 11, "page_number": 3, "is_center": false, "text": "Kitchen — All suites feature…", "stable_citation": "…" },
    { "chunk_id": 84213, "chunk_index": 12, "page_number": 3, "is_center": true,  "text": "…integrated Miele appliance package including a 30 inch gas cooktop…", "stable_citation": "…" },
    { "chunk_id": 84214, "chunk_index": 13, "page_number": 3, "is_center": false, "text": "…panel-ready dishwasher and a stacked washer/dryer.", "stable_citation": "…" }
  ],
  "meta": {
    "chunk_id": 84213,
    "document_id": 1422,
    "project_id": 6809,
    "extraction_id": 5567,
    "before_returned": 1,
    "after_returned": 1
  }
}

📘
Chunk IDs are tied to a specific extraction. If a document is re-indexed, an
old chunk_id may return "not found" — re-run search-document-text to get
fresh IDs.

A typical flow

find-projects-mentioning → "which projects have Miele appliances?"
search-document-text (with project_id) → exact chunks for a project.
fetch-document-context (with a chunk_id) → surrounding paragraphs to
confirm the detail before presenting it.

These are MCP tools, invoked by an MCP client (e.g. Claude) connected to

Availability

How matching works

Categories

Coverage metadata

search-document-text

Input

Output

find-projects-mentioning

Input

Output

list-project-documents

Input

Output

fetch-document-context

Input

Output

Chunk IDs are tied to a specific extraction. If a document is re-indexed, an

A typical flow

`search-document-text`

`find-projects-mentioning`

`list-project-documents`

`fetch-document-context`