# ra-mcp-search
Search domain package for Riksarkivet transcribed documents.
## Overview
Contains the business logic for searching the Swedish National Archives. This is a pure domain package — no MCP or CLI dependency. It provides Pydantic models, an API client, and a search operations layer used by both `ra-mcp-search-mcp` (MCP tools) and `ra-mcp-search-cli` (CLI command).
## Components
- **models.py**: Pydantic models — `SearchRecord` (single result with metadata + transcribed text snippets), `RecordsResponse` (API response wrapper), `SearchResult` (paginated result set)
- **clients/search_client.py**: `SearchAPI` — client for the Riksarkivet records API (`https://data.riksarkivet.se/api/records`)
- **operations/search_operations.py**: `SearchOperations` — high-level search orchestration with pagination, filtering, and dedup support
- **config.py**: API URL, default limit (25), and max display (10) constants
## Usage
```python
from ra_mcp_common.utils.http_client import HTTPClient
from ra_mcp_search.search_operations import SearchOperations
ops = SearchOperations(http_client=HTTPClient())
result = ops.search(
keyword="trolldom",
transcribed_only=True,
only_digitised=True,
offset=0,
limit=25,
sort="relevance",
year_min=1600,
year_max=1700,
)
for record in result.items:
print(record.metadata.reference_code, record.metadata.title)
if record.transcribed_text:
for snippet in record.transcribed_text.snippets:
print(f" Page {snippet.pages[0].id}: {snippet.text[:100]}...")
```
## Search Modes
- **Transcribed text** (`transcribed_only=True`): Full-text search across ~1.6M AI-transcribed pages (court records, 17th-18th century)
- **Metadata** (`transcribed_only=False`): Search 2M+ catalog records by title, name, place, description
## Dependencies
- Internal: `ra-mcp-common`
- External: `pydantic`
## Part of ra-mcp
Used by `ra-mcp-search-mcp` (MCP tools) and `ra-mcp-search-cli` (CLI command). See the [docs site](https://ai-riksarkivet.github.io/ra-mcp/) for full project documentation.
