MCP 服务器

DocImprint Document Intelligence

Verifiable document intelligence for AI agents. Extract text, tables, and structured data from PDFs and URLs. Summarize, answer questions, check claims, and translate — all with cited evidence. Store tamper-evident evidence bundles with cryptographic signatures and on-chain attestation via Base L2. Cross-document semantic search and Q&A across named collections. Pay per call with USDC

README

Document memory agents can prove. PDFs & URLs → cited answers · evidence bundles · on-chain attestation

At a glance
Choose your integration
See what you get
How it works
Why not plain OCR or RAG?
Use cases
Install
Quick start
Features
Why DocImprint?
Agent-native features
CrewAI integration
MCP server
API reference
Error handling
Authentication
TypeScript
Pricing
Community
Links

At a glance

DocImprint turns any PDF or URL into a tamper-evident evidence bundle — structured data, AI-cited answers, and a cryptographic proof your agents can verify independently.


Input	PDF or URL
Output	`ev_...` evidence bundle — cited answers, artifact hashes, EIP-191 signature
Verify	`client.verify()` — free, no auth; optional Base L2 notarization via EAS

Choose your integration

Path	Best for	Start here
TypeScript SDK	Node agents, typed apps	Quick start
Python SDK	CrewAI, scripts	Quick start · CrewAI
MCP server	Claude, Cursor	MCP server
x402 USDC	Autonomous agents, no account	x402 USDC payments

See what you get

Every response includes a verifiable bundle ID, manifest hash, and cited answers tied to exact source quotes.

{
  "bundle_id": "ev_01jqv8k3m2x",
  "manifest_sha256": "a3f2c1d8e9b0476f8a2b1c0d9e8f7a6b5c4d3e2f1a0b9c8d7e6f5a4b3c2d1e0f",
  "signature": {
    "signature": "0x8f4e2a1b9c3d5e7f0a2b4c6d8e0f1a3b5c7d9e1f3a5b7c9d1e3f5a7b9c1d3e5f71c",
    "signer_address": "0x742d35Cc6634C0532925a3b844Bc9e7595f0EbE",
    "algorithm": "secp256k1-eip191"
  },
  "answer_cited": {
    "value": "Section 4.2 does not permit unilateral termination without 90 days written notice.",
    "citations": [
      {
        "quote": "Neither party may terminate this Agreement unilaterally except upon ninety (90) days prior written notice to the other party.",
        "paragraphs": [42],
        "page": 4,
        "confidence": "high"
      }
    ]
  },
  "artifacts": {
    "markdown": { "sha256": "b4e5f6a79876543210fedcba9876543210fedcba9876543210fedcba9876543210ab" },
    "manifest": { "sha256": "a3f2c1d8e9b0476f8a2b1c0d9e8f7a6b5c4d3e2f1a0b9c8d7e6f5a4b3c2d1e0f" }
  }
}

How it works

flowchart LR
  source[PDF or URL] --> extract[extract / qa / checkClaims]
  extract --> bundle[Evidence bundle ev_...]
  bundle --> artifacts[Artifacts + manifest SHA-256]
  bundle --> citations[Cited answers with quotes]
  artifacts --> verify[verify - free]
  verify --> notarize[notarize on Base L2 - optional]

Why not plain OCR or RAG?

	Plain OCR	Vector RAG	DocImprint
Exact source quotes	No	Approximate	Yes, with paragraph refs
Tamper evidence	No	No	SHA-256 manifest + EIP-191 sig
Independent verify	No	No	`verify()` without API key
On-chain attestation	No	No	EAS on Base L2
Agent audit trail	No	No	Provenance + handoff (Python)

Use cases

Scenario	Method	Result
Contract Q&A	`client.qa()`	Cited answer with confidence score
Claim verification	`client.checkClaims()`	Per-claim verdict: `supported` / `contradicted` / `not_found`
Invoice intake	`client.extract({ mode: 'invoice' })`	Structured merchant, line items, totals — all cited
Compliance archive	`legal_hold: true` + `notarize()`	Immutable on-chain record, deletion blocked

Install

npm install docimprint           # TypeScript / Node.js

pip install docimprint           # Python REST client
pip install "docimprint[crewai]" # + 10 CrewAI tools & ProvenanceTracker

Python source lives in python/. PyPI: docimprint.

Quick start

Get an API key at docimprint.com — free tier available, no credit card required.

TypeScript

import { DocImprintClient } from 'docimprint'

const client = new DocImprintClient({ apiKey: 'dr_live_...' })

const result = await client.extract({
  source: 'https://example.com/contract.pdf',
  include: ['markdown', 'summary'],
})

console.log(result.bundle_id)        // ev_01j...
console.log(result.summary)          // AI-generated summary
console.log(result.manifest_sha256)  // tamper-evident hash
console.log(result.key_points_cited[0].citations[0].quote)

Python

from docimprint import DocImprintClient

client = DocImprintClient(api_key="dr_live_...")

result = client.extract(
    url="https://example.com/contract.pdf",
    include=["markdown", "summary"],
)
print(result["bundle_id"])         # ev_01j...
print(result["summary"])           # AI-generated summary
print(result["manifest_sha256"])   # tamper-evident hash

REST API (cURL)

curl -X POST https://api.docimprint.com/v1/extract \
  -H "Authorization: Bearer dr_live_..." \
  -H "Content-Type: application/json" \
  -d '{"source": "https://example.com/contract.pdf", "include": ["markdown", "summary"]}'

Features

Capability	Method / Mode	Description
Extract	`extract()`	Full evidence bundle — markdown, summary, cited key points, artifacts, manifest SHA-256
Invoice parsing	`mode: 'invoice'`	Structured `InvoiceResult`: merchant, date, line items, subtotal, tax, total — all with citations
Document comparison	`mode: 'compare'`	Diff two documents; supply `previous_bundle_id` to track changes
Structured extraction	`mode: 'extract-structured'`	Extract typed fields using a caller-defined JSON Schema
Summarize	`summarize()`	Prose summary + key points, each with cited paragraph references
Q&A	`qa()`	Single-question answer with inline citations and confidence score
Claim-check	`checkClaims()`	Per-claim verdict: `supported` / `contradicted` / `not_found` with evidence quotes
Translate	`translate()`	Source-cited translation to any language
Describe	`describe()`	AI description of image or PDF page contents
Collections	`createCollection()` + `searchCollection()` + `askCollection()`	Semantic search + cross-document RAG Q&A across multiple bundles
Verify	`verify()`	Cryptographic integrity check — manifest hash, EIP-191 signature, artifact hashes. Free, no auth
Notarize	`notarize()`	Write manifest SHA-256 to Base L2 via EAS — permanent, immutable attestation
MCP server	20 tools · 2 prompts	Native MCP for Claude, Cursor, and any MCP-compatible client
x402	REST `X-Payment` header	Pay per call in USDC — no account required

Why DocImprint?

🔏 Cryptographic provenance — every bundle is EIP-191 signed at creation; optionally notarized on Base L2 via EAS for an immutable on-chain audit trail
🤖 Agent-native by design — async jobs, webhooks, idempotency keys, and legal hold built into every request, not bolted on
📎 Beyond OCR — citations carry exact quotes, paragraph indices, and confidence scores (high/medium/low), not just raw text
🛠️ 10 CrewAI tools out of the box — research_tools(), legal_tools(), collection_tools() pre-grouped for common agent workflows
⚖️ Compliance-ready — legal_hold, provenance logging, multi-agent handoff tracking, and on-chain notarization designed for legal and regulated industries
💳 Flexible payment — monthly credit plans or pay-per-call USDC via x402, no account required

Agent-native features

DocImprint is designed for autonomous agent workflows, not just synchronous API calls.

// Fire-and-forget async extraction — returns job_id immediately
const job = await client.extract({
  source: 'https://example.com/large-report.pdf',
  async: true,
  webhook: 'https://your-agent.io/callback',
  idempotency_key: 'report-2025-q4',
  legal_hold: true,
})

const status = await client.getJob(job.job_id)
// { status: 'complete', bundle_id: 'ev_...', progress_pct: 100 }

// Monitor a URL for changes — get notified on diff
await client.extract({
  source: 'https://example.com/terms.html',
  monitor: { webhook: 'https://your-agent.io/changes', mode: 'diff' },
})

Python: provenance & multi-agent handoff tracking

client.log_provenance(bundle_id="ev_...", agent_id="agent-research", action="extracted")
client.handoff(bundle_id="ev_...", from_agent="agent-research", to_agent="agent-legal", note="ready for claim check")

CrewAI integration

10 purpose-built tools for CrewAI agents, organized into preset groups.

from docimprint.crewai import DocImprintToolkit

toolkit = DocImprintToolkit(
    api_key="dr_live_...",
    collection_id="col_...",  # required for collection tools
)

toolkit.research_tools()      # extract, summarize, qa, check_claims
toolkit.legal_tools()         # check_claims, verify, notarize
toolkit.collection_tools()    # search, ask, add_to_collection
toolkit.all_tools()           # all 10 tools

Tool	What it does
`ExtractEvidenceTool`	Full evidence bundle with citations and manifest hash
`SummarizeTool`	Key points with cited paragraph references
`QATool`	Cited answer with confidence score
`CheckClaimsTool`	Per-claim verdict: supported / contradicted / not_found
`TranslateTool`	Source-cited translation
`VerifyBundleTool`	Cryptographic integrity check — manifest, hash, signature
`NotarizeTool`	On-chain EAS attestation on Base L2
`SearchCollectionTool`	Semantic vector search across document collection
`AskCollectionTool`	Cross-document RAG Q&A with cited sources
`AddToCollectionTool`	Add bundle to collection and trigger async indexing

ProvenanceTracker wraps all tools to automatically log agent actions and bundle handoffs.

from docimprint.crewai import DocImprintToolkit, ProvenanceTracker

tracker = ProvenanceTracker(client=toolkit.client)
trackable = toolkit.trackable_tools()

DocImprintKnowledgeSource integrates with CrewAI's knowledge system for retrieval-augmented agents.

MCP server

DocImprint exposes a native MCP server for use with Claude, Cursor, and any MCP-compatible client.

Install via Smithery (one command):

npx @smithery/cli install docimprint --client claude

Manual config:

{
  "mcpServers": {
    "docimprint": {
      "type": "streamable-http",
      "url": "https://api.docimprint.com/mcp",
      "headers": {
        "Authorization": "Bearer dr_live_..."
      }
    }
  }
}

Transport: streamable-http · Auth: Bearer token · Listed on Smithery · Listed on Glama

MCP tools

URL tools — lean mode, no bundle stored:

Tool	Description
`extract_url`	Fetch URL → text + metadata
`summarize_url`	Fetch URL → prose summary + key points
`qa_url`	Fetch URL → cited answer to a specific question
`translate_url`	Fetch URL → translated content

Document tools — accepts base64 PDF or image:

Tool	Description
`extract_text`	OCR plain text from PDF / image
`extract_tables`	OCR tables as Markdown from PDF / image
`parse_invoice`	Structured invoice fields: merchant, line items, totals + citations
`summarize_document`	Prose summary + cited key points
`check_claims`	Per-claim verdict with evidence quotes
`extract_structured`	Extract typed fields using a caller-defined schema

Bundle & collections:

Tool	Description
`verify_bundle`	Cryptographic integrity check
`get_bundle`	Bundle metadata and notarization status
`notarize_bundle`	On-chain EAS attestation on Base L2
`create_collection`	Create a named document collection
`list_collections`	List your collections
`add_document_to_collection`	Add bundle + trigger async indexing
`search_collection`	Semantic vector search
`ask_collection`	Cross-document RAG Q&A
`get_job_status`	Poll async job (extract, indexing, batch)
`get_quota`	Check credit balance and plan status

Guided prompts: claim_check_workflow · invoice_intake

Resource: bundle://{bundle_id} — read bundle metadata directly

API reference

Full typed reference: docimprint.com/docs · OpenAPI

Category	Methods
Core	`extract`, `verify`, `download`, `notarize`, `deleteBundle`
Focused	`summarize`, `qa`, `translate`, `checkClaims`, `describe`
Extract modes	`invoice`, `extract-structured`, `compare`, `store: false` (lean)
Collections	`createCollection`, `listCollections`, `addToCollection`, `searchCollection`, `askCollection`
Jobs	`getJob`, `listJobs`, `getQuota`

Error handling

import { DocImprintClient, DocImprintError } from 'docimprint'

try {
  const result = await client.extract({ source: 'https://example.com/doc.pdf' })
} catch (err) {
  if (err instanceof DocImprintError) {
    console.error(err.message)    // human-readable error
    console.error(err.status)     // HTTP status code
    console.error(err.requestId)  // x-request-id for support
  }
}

Authentication

API key

Monthly credits via Stripe — sign up at docimprint.com, free tier available:

const client = new DocImprintClient({ apiKey: 'dr_live_...' })

x402 USDC payments

DocImprint supports the x402 open standard — pay per call in USDC on Base, straight from any EVM wallet. No account. No sign-up. No API key. Your agent can call the API autonomously without any human-managed credentials.

The API is x402-native: it returns a standard 402 Payment Required response with on-chain payment details, your client settles in USDC, and the request completes automatically. From $0.01 / call.

With @x402/fetch (automatic payment handling):

import { wrapFetchWithPayment } from '@x402/fetch'
import { createWalletClient, http } from 'viem'
import { privateKeyToAccount } from 'viem/accounts'
import { base } from 'viem/chains'

const account = privateKeyToAccount('0x...')
const wallet = createWalletClient({ account, chain: base, transport: http() })

const fetchWithPayment = wrapFetchWithPayment(fetch, wallet)

const res = await fetchWithPayment('https://api.docimprint.com/v1/summarize', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ source: 'https://example.com/report.pdf' }),
})
const data = await res.json()

Identity on owner routes: For GET/DELETE operations that require ownership (e.g. fetching your bundle), pass the payment-signature header from a prior payment response so the API can verify your wallet address without a separate login.

TypeScript

All request and response types are exported:

import type {
  ExtractRequest,
  ExtractResponse,
  Citation,
  CitedField,
  InvoiceResult,
  Job,
  Collection,
  SearchResult,
} from 'docimprint'

Pricing

Free tier available — no credit card required. Monthly credit plans via Stripe, or pay per call in USDC via x402. See pricing →

Community

Questions, integrations, and announcements: GitHub Discussions.