Verifiable document memory for MCP agents

Ready to connect? The how-to lives in docs — key, host config, first tool call, verify.

Also listed on Smithery .

Why does verifiability matter for MCP users?

MCP hosts invoke tools on behalf of users. Without bundle_id and manifest_sha256, an agent cannot prove which document version it read. DocImprint MCP tools return the same signed evidence envelope as REST — so hosts can audit, cite, and verify agent outputs.

What MCP tools does DocImprint expose?

extract_text, extract_tables, parse_invoice, check_claims, extract_structured, verify_bundle, search_collection, and ask_collection — all returning citation-grounded results with optional Merkle proofs when store=true.

How do I connect Claude Desktop to DocImprint MCP?

Follow the Agent Quickstart at /docs/agents: use the mcp-remote stdio bridge with Authorization: Bearer dr_live_… pointing to https://api.docimprint.com/mcp, then run extract_text and verify_bundle.

Does MCP support offline verification?

Yes. Download the evidence bundle ZIP, recompute manifest SHA-256, verify the platform secp256k1 signature against GET /v1/keys — same flow as REST. verify_bundle runs server-side checks; offline verify needs no ongoing API trust.

Tools at a glance

Full setup steps are in the Agent Quickstart.

jsonClaude Desktop preview

{
  "mcpServers": {
    "docimprint": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.docimprint.com/mcp",
        "--header",
        "Authorization:${DOCIMPRINT_AUTH}"
      ],
      "env": {
        "DOCIMPRINT_AUTH": "Bearer dr_live_YOUR_KEY"
      }
    }
  }
}

Preview only — copy the complete configs and verify steps from docs.

Document

document.extract_textPDF or image → raw text via OCR
document.extract_tablesPDF or image → Markdown tables
document.parse_invoiceInvoice/receipt → merchant, line items, totals
document.check_claimsDocument + claims[] → supported / contradicted / not_found
document.extract_structuredDocument + JSON schema → typed fields with citations
document.summarizeDocument → summary + key_points[]

URL

url.extractHTTPS URL → page text
url.summarizeHTTPS URL → summary + key_points[]
url.qaHTTPS URL + question → cited answer
url.translateHTTPS URL + target language → translated text

Bundle

bundle.verifybundle_id → integrity check (returns receipt)
bundle.getbundle_id → metadata + artifacts (returns receipt)
bundle.notarizeAnchor manifest hash on Base L2 (returns receipt)
receipt.listbundle_id → signed action receipts[]
receipt.verifyrcpt_… → signature + manifest binding check

Collections

collection.createCreate a named document corpus
collection.listList collections you own
collection.add_documentIndex a bundle into a collection (returns receipt)
collection.searchSemantic search over indexed chunks
collection.askCross-document Q&A with citations

Ops

job.statusPoll async job by job_id
account.quotaCredits used / remaining (free)

Agent Quickstart

Config, first tool, verify

For agents

Discovery & marketplace copy

Document memory

Concept reference