MCP documentation

Auto-generated from service modules. Agents should read the /llms.txt index.

Canonical payloads Payloads markdown

• Hello — Smoke-test MCP service (hello world) and host Server status. (markdown · payload)
• Geo — GeoNames-backed geospatial MCP: countries, cities, admin regions, marine, POIs (airports, peaks, lakes, …), distance and nearby search. (markdown · payload)
• Plant — Search and resolve plant taxa from a local World Flora Online backbone (scientific names, synonyms). No live external plant APIs. (markdown · payload)
• Herbal — Search herbal preparations by condition/symptom from local multi-source data (EMA HMPC, Dr. Duke ethnobotany, Wikidata, WHO/NCCIH seeds) linked to WFO plant ids, plus dump/curated active constituent profiles. Query language: English MeSH headings or short HU symptom aliases for conditions; English/INN names for constituents. Not medical advice. No live external APIs. (markdown · payload)
• Compound — Central hatóanyag / active-ingredient master (compound_id). Resolve names, lookup cards, list herbal plants containing a compound. Not medical advice. No live external chemistry APIs. (markdown · payload)
• Supplement — Search dietary supplement products from a local NIH DSLD label dump (~200k US products). Ingredients link to the central compound master. Label data — not lab assays. Not medical advice. No live NIH API. (markdown · payload)
• Drug — Search drugs and actives from local dumps (DrugCentral, ChEMBL, DailyMed, FDA NDC/Drugs@FDA, UNII). Ingredients resolve to the central compound master. Query language: English/INN and US/EU brand names; indications in English (diabetes, not cukorbetegség). Not medical advice. No live regulatory APIs on the request path. (markdown · payload)
• Fetch — Fetch a public URL and return its main content as markdown (Readability + markdownify, same approach as the official MCP fetch server). (markdown · payload)
• Search — Agent web research: search_web/search_news return links+snippets only; then open 1–2 HTML URLs with search_fetch or search_fetch_summary. Default locale us/en (Hungarian: gl=hu, hl=hu). site is 1–5 hostnames. PDFs → docs service. (markdown · payload)
• Docs — Extract text and tables from PDF, DOCX, XLSX (and plain text/CSV). Scanned PDFs use Tesseract OCR (hun/eng/deu). (markdown · payload)
• Shot — Render a public URL to a PNG/JPEG screenshot (Playwright + system Chromium). (markdown · payload)
• Image — Resize and convert images (JPEG/PNG/WebP) to a staging URL. (markdown · payload)
• Email — Validate and inspect email addresses (syntax, MX via dig, disposable, role). Does not send mail. (markdown · payload)
• QR — Create and read QR codes and 1D barcodes; vCard / Wi-Fi / EPC payment QR helpers. Does not initiate payments. (markdown · payload)
• PDF — Convert Markdown or HTML to a PDF or DOCX published as a temporary staging URL. (markdown · payload)
• Validate — Offline validators for JSON, JSON Schema, XML, CSV, URL, IBAN, EU VAT (checksum), and Hungarian tax/company numbers. No email (use email service); no VIES. (markdown · payload)
• Promptguard — Host-side prompt injection gate (normalize + structural + intent + soft neural stack: attack, discourse, gated secondary semantic encoder + policy; optional sticky session risk, spotlight/canary). Call from orchestrator BEFORE every main LLM call — never via model tool-choice. Returns injection bool + score; caller applies policy. Pass session_id for multi-turn sticky risk. Detection is imperfect — enforce tool allowlists, least privilege, and human approval for dangerous ops. (markdown · payload)
• Whisper — Speech-to-text (Whisper large-v3 CPU, default hu). Async: staging upload → whisper_job_start → whisper_job_status → whisper_job_result. Large file: POST /staging/upload (Bearer, multipart file) → pass response.url to whisper_job_start. (markdown · payload)
• Staging — Temporary public HTTPS URL for large files (max 1 GiB, ~24h TTL). Upload: HTTP multipart POST /staging/upload, staging_put_text (UTF-8, no base64), staging_put_base64 (small binaries), or staging_ingest_url. (markdown · payload)
• GLC-TTS — Hungarian text-to-speech (GLC-TTS, CPU, zero-shot voice clone). Async: tts_list_voices → tts_job_start → tts_job_status → tts_job_result. Output WAV via staging URL. ref_text via platform Whisper when transcribe_ref=true. (markdown · payload)
• Vision — Image analysis with Ternary Bonsai 27B via Together AI (English JSON). Async: POST /vision/upload or /staging/upload → vision_job_start({ image_url }) → status → result. (markdown · payload)
• Weather — Open-Meteo weather MCP: current, 16-day forecast, ERA5 history since 1940, air quality/pollen, marine (Adriatic / lakes). Place names via GeoNames. 1 credit per successful data call. (markdown · payload)
• Embed — Local multilingual MiniLM embeddings (384-d). Same model the host Signal Engine already uses. 1 credit per 1 million accrued tokens (typical call is 0). (markdown · payload)

curl -sS -X POST 'https://mcp.glc-rag.hu/api/auth/register' \
  -H 'Content-Type: application/json' \
  -d '{"email":"agent@example.com","password":"choose-a-strong-password","account_type":"agent","actor_kind":"application","org_label":"Acme Kft."}'

URL: https://mcp.glc-rag.hu/mcp
Header: Authorization: Bearer <api-token>