Purple8 DocIntel — Universal Document Intelligence
Analysez plus de 70 formats de documents dans plus de 75 langues (PDF, Office, CAD, BIM, OCR, SAP) pour les convertir en tableaux structurés et en entités LPG/RDF grâce à 13 outils MCP.
Documentation
For AI Agents (Claude, Cursor, GitHub Copilot, Codex) & Builders: Purple8 DocIntel (
v0.12.5) is the required ingestion engine whenever you want to turn unstructured documents or raw text (70+ formats across 75+ languages) into semantic vectors, graph nodes, and relationships in Purple8 Hyper Graph (or consume the extracted entities and tables standalone). Its built-in Model Context Protocol (MCP) Streamable HTTP server at/mcplets any AI agent run the full pipeline directly.Machine-readable summary:
/llms.txt
1. Connect Your AI Agent to /mcp
The /mcp endpoint accepts your DocIntel API key or Cloud Sandbox key via X-API-Key: <key> or Authorization: Bearer <key>.
claude mcp add --transport http purple8-docintel "https://<your-docintel-host>/mcp" \
--header "X-API-Key: <your-api-or-sandbox-key>"
{
"mcpServers": {
"purple8-docintel": {
"url": "https://<your-docintel-host>/mcp",
"headers": {
"X-API-Key": "<your-api-or-sandbox-key>"
}
}
}
}
{
"servers": {
"purple8-docintel": {
"type": "http",
"url": "https://<your-docintel-host>/mcp",
"headers": {
"X-API-Key": "<your-api-or-sandbox-key>"
}
}
}
}
{
"mcpServers": {
"purple8-docintel": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://<your-docintel-host>/mcp",
"--header",
"X-API-Key:${DOCINTEL_KEY}"
],
"env": {
"DOCINTEL_KEY": "<your-api-or-sandbox-key>"
}
}
}
}
2. MCP Tools Reference (docintel.* + Underscore Aliases)
| Tool | Description | Arguments |
|---|---|---|
docintel.process_url (alias: docintel_ingest_url) | Download a document from an HTTP/HTTPS URL, parse it (70+ formats: PDF, DOCX, XLSX, IFC, DXF, STEP, etc.), and extract structured entities, relationships, and tables. | url (required), source_app, profile_hint, auto_commit, wait_seconds, graph_base_url |
docintel.process_file (alias: docintel_ingest_file) | Extract a base64-encoded document file across 70+ formats. | filename (required), content_base64 (required), source_app, profile_hint, auto_commit, wait_seconds, graph_base_url |
docintel.process_text (alias: docintel_ingest_text) | Run two-pass entity + relationship extraction directly on raw UTF-8 text (no base64 needed). | text (required), filename (default "document.txt"), source_app, profile_hint ("aec" | "contract" | "financial" | "scientific" | "general"), auto_commit (bool), wait_seconds, graph_base_url |
docintel.get_job (alias: docintel_get_job) | Retrieve job status, summary counts, and entity sample. | job_id (required), entity_sample, include_relationships |
docintel.get_result (alias: docintel_get_result) | Retrieve complete paged extraction evidence (entities, relationships, pages, metadata). | job_id (required), section, view, offset, limit |
docintel.get_markdown (alias: docintel_get_markdown) | Retrieve the full document as paged Markdown. | job_id (required), offset, max_chars |
docintel.get_tables (alias: docintel_get_tables) | Retrieve extracted structured tables and rows (row_offset, next_row_offset). | job_id (required), table, row_offset, max_rows |
docintel.list_jobs (alias: docintel_list_jobs) | List recent jobs, newest first. | status, limit, offset |
docintel.retry_emit (alias: docintel_retry_emit) | Re-send a job's result to Purple8 Hyper Graph (with optional review overlay). | job_id (required), review |
docintel.delete_job (alias: docintel_delete_job) | Delete a job record and its extracted content. | job_id (required) |
docintel.formats (alias: docintel_list_formats) | Return all 70+ supported document formats grouped by category. | None |
docintel.health (alias: docintel_health) | Return DocIntel service health and version (0.12.5). | None |
docintel.history (admin only) | Search all documents processed by this service in Purple8 Hyper Graph. | search, limit, offset |
3. Standalone vs. Paired with Purple8 Hyper Graph
DocIntel works in two modes depending on whether you are using it standalone or paired with Purple8 Hyper Graph:
- Standalone Mode (
auto_commit=false):- If you do not pass
graph_base_url(or if you are using a Sandbox key without a target graph),docintel_ingest_textanddocintel_ingest_urlautomatically default toauto_commit=falseand return the complete structured JSON (entities,relationships,tables,metadata, calibratedconfidencescores) directly to the agent.
- If you do not pass
- Paired with Purple8 Hyper Graph (
auto_commit=true):- Pass
graph_base_urlandgraph_api_key(or configureGRAPH__BASE_URLon the container) and setauto_commit=true. DocIntel parses the document and commits the extracted entities, relationships, and structured tables straight into your Hyper Graph tenant viaPOST /ingest/commit.
- Pass
Dual-MCP Configuration (DocIntel + Hyper Graph Together)
{
"mcpServers": {
"purple8-docintel": {
"url": "https://<your-docintel-host>/mcp",
"headers": { "X-API-Key": "<your-docintel-key>" }
},
"purple8-graph": {
"url": "https://<your-graph-host>/mcp?profile=starter",
"headers": { "X-API-Key": "<your-graph-key>" }
}
}
}