Labsco
Kos-M logo

ACG Mcp

โ˜… 1

from Kos-M

Standalone MCP server for the Audited Context Generation (ACG) Protocol โ€” verifiable fact-checking and grounded RAG via MongoDB.

๐Ÿ”ฅ๐Ÿ”ฅ๐Ÿ”ฅโœ“ VerifiedFreeAdvanced setup

ACG MCP Server

Standalone MCP server for the Audited Context Generation (ACG) Protocol โ€” verifiable fact-checking and grounded RAG via MongoDB.

ACG provides a dual-layer standard for veracity assurance:

  • UGVP (Layer 1): Atomic fact grounding with Claim Markers and Source Hash Identity (SHI)
  • RSVP (Layer 2): Logical synthesis verification with Relationship Markers

Features

  • Index URLs โ†’ Extract text, chunk by sentences, generate embeddings, store in MongoDB
  • Search Sources โ†’ Semantic (vector) + keyword search across indexed content
  • Check Indexed โ†’ Confidence-scored lookup to avoid unnecessary web_fetch calls
  • Generate Grounded Text โ†’ Create verifiable output with inline Claim Markers
  • Verify Claims โ†’ Re-fetch sources, fuzzy-match claims against source text
  • Build VAR โ†’ Generate machine-readable Veracity Audit Registry (SSR + RAR)
  • Crawl & Index โ†’ BFS URL discovery + automatic ACG indexing pipeline
  • Reset Database โ†’ Drop all ACG collections (with confirmation guard)

Available Tools

ToolDescription
acg_index_urlIndex a URL for ACG โ€” fetches, chunks, embeds, stores
acg_check_indexedCheck if a query has results in indexed sources
acg_search_sourcesSearch indexed sources by keyword
acg_list_sourcesList all indexed sources
acg_count_sourcesCount total indexed sources
acg_generate_grounded_textCreate text with Claim Markers (UGVP)
acg_verify_claimsVerify claims against their sources (fuzzy matching)
acg_build_varBuild Veracity Audit Registry (SSR + RAR)
acg_crawl_and_indexCrawl + index multiple URLs (background support)
acg_crawl_statusCheck background crawl task status
acg_crawl_list_tasksList all background crawl tasks
acg_reset_databaseโš ๏ธ Delete all indexed data (requires confirm=true)

Database Collections

The server uses a standard MongoDB collection structure:

CollectionPurpose
sourcesSource metadata (url, shi_prefix, url_hash, total_chunks)
dataChunks with embeddings (source_id, text, sentences, embedding)
claimsVerified claims (claim_id, shi_prefix, claim_text, verified)
relationshipsRSVP relationship records (rel_id, rel_type, claim_ids)
var_entriesVeracity Audit Registry entries

Indexes are auto-created on first connection.