mcpdirectory.dev
9,129 servers · updated 23 Sept

Footnote MCP server

official from Brick Byte Active last commit 40 d
{
  "mcpServers": {
    "brick-byte-footnote": {
      "command": "npx",
      "args": [
        "-y",
        "@brick-byte/footnote@latest"
      ],
      "env": {
        "DATABASE_URL": "postgresql://user:password@host:5432/db"
      }
    }
  }
}
file .mcp.json (in your project root)install npx -y @brick-byte/footnoteversion 0.1.1

What it does

Footnote is an MCP server that ensures every number in an AI answer traces back to specific rows in a business database, with tamper-evident provenance records. It is for organizations that need AI to report on financial, inventory, or procurement data where accuracy and auditability are required.

  • Query a business database through natural language and receive answers where every monetary figure or key value is clickable and links back to the exact rows that produced it.
  • Export an AI-generated report with full provenance chains so auditors can verify that numbers came from the system of record, not from model hallucination.
  • Refuse to answer questions that do not map to a valid database operation, rather than having the model invent an answer to an unroutable query.

Tools this server exposes

ToolDoes
operation_executionAccepts a chosen operation name and typed arguments, validates them against the database, executes the query, and returns structured results with evidence ident
provenance_appendRecords a claim-to-evidence mapping server-side with the question, answer, operations executed, and evidence fields, creating a hash-chained log.
entity_validationResolves customer names, invoice numbers, or other entities against the database before any query runs, returning a recoverable error with alternatives if the e
chain_verificationWalks the provenance log and detects any record that was modified after it was written by checking hash chains.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.