NDA Sentry

Anonymous NDA risk analysis for AI agents. $9 per report. No signup, no account, no data retention.

NDASentry exposes a multi-stage NDA risk analysis pipeline as MCP tools so any AI agent (Claude Desktop, Cursor, Cline, etc.) can review NDAs on behalf of its user — discover risky clauses, flag missing protections, return a structured risk report — without asking the user to leave the agent, create an account, or wait for a lawyer.

The wedge: every other legal MCP server in this space requires an account, an enterprise login, or attorney-in-the-loop review. NDASentry is designed to be the simplest path for a personal AI agent: call anonymously, pay $9 with a card, get a structured analysis in under a minute.

What it does

Two tools:

• preview_nda_risk(pdf_base64, filename) — Free preview. Stages the NDA, runs cheap-stage detection (regex-based clause finding, no LLM calls), returns a clause-level summary plus a Stripe payment link. Safe to expose to anonymous agent traffic; zero LLM cost on the preview path.

• get_nda_report(session_token) — Paid full report. Polls payment status, then runs the full multi-stage LLM pipeline (qualifier, detector, scorer, critic, synthesizer, decision policy) and returns structured JSON with clause-level risk findings, aggressive-clause signals, missing protections, a critique, and a recommended action.

The full pipeline output is designed for agent consumption — flat structured JSON, every clause carries risk level, evidence, and reasoning, so agents can filter, summarize, or route based on what their user cares about.

Quick start (hosted)

Point your MCP client at the public hosted endpoint:

https://nda-mcp-production.up.railway.app/mcp

Claude Desktop config

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "ndasentry": {
      "url": "https://nda-mcp-production.up.railway.app/mcp",
      "transport": "streamable-http"
    }
  }
}

Restart Claude Desktop. The two tools will appear in the agent's tool catalog.

Example agent prompt

Review this NDA and tell me if there is anything I should push back on before signing. [attach NDA PDF]

The agent calls preview_nda_risk first, returns a preview plus a payment URL. You pay $9 in your browser. The agent calls get_nda_report and returns the full analysis.

Self-host

For users who want their own instance, or local development:

git clone https://github.com/valtirman/ndasentry-mcp.git
cd ndasentry-mcp/mcp_server
python -m venv .venv-mcp
source .venv-mcp/bin/activate
pip install -r requirements.txt

export NDASENTRY_BACKEND_URL=https://ndasentry.ai
python -m mcp_server.server

The server listens on http://localhost:1966/mcp by default. Point your MCP client at it:

{
  "mcpServers": {
    "ndasentry": {
      "url": "http://localhost:1966/mcp",
      "transport": "streamable-http"
    }
  }
}

Configuration

Env var	Default	Purpose
NDASENTRY_BACKEND_URL	http://localhost:8001	Backend API that runs the analysis pipeline
PORT	1966	Port the MCP server binds to
MCP_ALLOWED_HOSTS	(empty)	Comma-separated production hosts to add to DNS rebinding allowlist
MCP_ALLOWED_ORIGINS	(empty)	Comma-separated production origins to add to DNS rebinding allowlist

The backend (ndasentry.ai by default for the hosted version) handles document analysis and Stripe payment verification. The MCP server is a thin protocol adapter that does not reimplement any pipeline logic. The same backend serves the web product at https://ndasentry.ai.

Tools reference

preview_nda_risk

Input:

• pdf_base64 (string): base64-encoded PDF bytes, max 10 MB
• filename (string): the PDF filename

Output: JSON with session_token, payment_url (Stripe link with the session token bound as client_reference_id), clause_summary, missing_required_clauses, a labeled sample clause showing the shape of a paid analysis, and a disclaimer.

get_nda_report

Input:

• session_token (string): from the preview_nda_risk response

Output: Full structured AnalysisReport JSON with clauses, risk scores, critique, completeness, qualification, aggressive-clause signals, recommended action, and disclaimer. If payment is not yet complete, returns a polling status response instead.

Disclaimer

NDASentry is a contract risk screening tool, not a law firm. Output is not legal advice and does not create an attorney-client relationship. For binding legal interpretation or high-stakes decisions, consult licensed counsel.

Privacy

The hosted backend at ndasentry.ai is designed around an "evaporates" model:

• Documents are processed in memory only, never written to disk
• Reports are cached in RAM for 5 minutes after generation, then deleted
• No account, no email collection, no document retention

About

Built by Val Tirman at FrontRange Mountain AI LLC. Solo indie effort. The web product runs at https://ndasentry.ai; this MCP server is the agent-facing channel on the same backend.

If you build something with NDASentry MCP, drop a note: [email protected]. Genuinely interested in what agents do with this.

License

MIT — see LICENSE file in the repository root.