document-evidence-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| PADDLE_OCR_BASE_DIR | No | Cache directory for old PaddleOCR models. | %USERPROFILE%\.paddleocr |
| PADDLE_PDX_CACHE_HOME | No | Cache directory for PP-StructureV3/PaddleX models. | %USERPROFILE%\.paddlex |
| DOCUMENT_EVIDENCE_STORE | No | Path to the document evidence store (SQLite and artifacts). | %LOCALAPPDATA%\document-evidence-mcp |
| PADDLE_PDX_MODEL_SOURCE | No | Model source: 'huggingface' or 'bos'. | huggingface |
| DOCUMENT_EVIDENCE_OCR_DEVICE | No | OCR device (e.g., 'gpu:0', 'cpu'). Empty for auto. | |
| DOCUMENT_EVIDENCE_CHUNK_CHARS | No | Target character count per evidence chunk. | 1600 |
| DOCUMENT_EVIDENCE_OCR_PROVIDER | No | OCR provider: 'none' or 'paddleocr'. | none |
| DOCUMENT_EVIDENCE_ALLOWED_ROOTS | No | Semicolon-separated list of allowed root directories for document import. | |
| DOCUMENT_EVIDENCE_CHUNK_OVERLAP | No | Overlap character count between chunks. | 160 |
| DOCUMENT_EVIDENCE_MAX_FILE_BYTES | No | Maximum file size in bytes for a single document. | 1073741824 |
| DOCUMENT_EVIDENCE_MAX_SEARCH_HITS | No | Maximum number of evidence hits per search. | 20 |
| DOCUMENT_EVIDENCE_MAX_SEARCH_CHARS | No | Hard limit on total response characters for a search. | 12000 |
| PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK | No | Set to 'True' to skip model source connectivity check. | False |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| ingest_documentA | Import a local document once and persist its structured evidence. The cache identity includes source SHA-256, parser configuration, OCR provider, languages and engine versions. Use force only to create an explicit new revision. |
| search_evidenceC | Search the persistent local index and return a bounded set of citations. |
| get_evidenceA | Read exact evidence blocks by stable ID under a total character budget. |
| list_documentsA | List recently indexed document versions without returning document bodies. |
| get_documentB | Return one document manifest summary and all source paths seen for it. |
| render_cropA | Render one PDF evidence rectangle to a local PNG artifact. Coordinates use PDF points (72 points per inch). This tool returns a path and metadata, never the image bytes. |
| doctorB | Report parser, SQLite/FTS and optional OCR availability. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 7 tools
Each tool targets a distinct aspect of the document evidence lifecycle: ingestion, searching, retrieval by ID, document listing, document metadata, rendering crops, and health checks. The overlapping-sounding pairs (search/get, list/get) are clearly differentiated by their descriptions and return types.
Six of seven tools follow a consistent verb_noun pattern (ingest_document, search_evidence, get_evidence, list_documents, get_document, render_crop). The lone exception is 'doctor', which breaks the pattern and is the only naming inconsistency.
Seven tools is well-scoped for a document evidence management server, covering the full workflow from ingestion through search, retrieval, rendering, and diagnostics. Each tool has a clear purpose with no redundancy.
The toolset provides a complete lifecycle from ingestion to evidence retrieval and rendering, including document management and a health check. There are no obvious missing operations, as updates can be handled via explicit new revisions during ingestion.