{
  "markdown": "# NDASentry MCP Server\n\n> Anonymous NDA risk analysis for AI agents. $9 per report. No signup, no account, no data retention.\n\nNDASentry 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.\n\nThe 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.\n\n## What it does\n\nTwo tools:\n\n- **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.\n\n- **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.\n\nThe 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.\n\n## Quick start (hosted)\n\nPoint your MCP client at the public hosted endpoint:\n\n    https://nda-mcp-production.up.railway.app/mcp\n\n### Claude Desktop config\n\nAdd to claude_desktop_config.json:\n\n    {\n      \"mcpServers\": {\n        \"ndasentry\": {\n          \"url\": \"https://nda-mcp-production.up.railway.app/mcp\",\n          \"transport\": \"streamable-http\"\n        }\n      }\n    }\n\nRestart Claude Desktop. The two tools will appear in the agent's tool catalog.\n\n### Example agent prompt\n\n> Review this NDA and tell me if there is anything I should push back on before signing.\n> [attach NDA PDF]\n\nThe 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.\n\n## Self-host\n\nFor users who want their own instance, or local development:\n\n    git clone https://github.com/valtirman/ndasentry-mcp.git\n    cd ndasentry-mcp/mcp_server\n    python -m venv .venv-mcp\n    source .venv-mcp/bin/activate\n    pip install -r requirements.txt\n\n    export NDASENTRY_BACKEND_URL=https://ndasentry.ai\n    python -m mcp_server.server\n\nThe server listens on http://localhost:1966/mcp by default. Point your MCP client at it:\n\n    {\n      \"mcpServers\": {\n        \"ndasentry\": {\n          \"url\": \"http://localhost:1966/mcp\",\n          \"transport\": \"streamable-http\"\n        }\n      }\n    }\n\n### Configuration\n\n| Env var | Default | Purpose |\n|---|---|---|\n| NDASENTRY_BACKEND_URL | http://localhost:8001 | Backend API that runs the analysis pipeline |\n| PORT | 1966 | Port the MCP server binds to |\n| MCP_ALLOWED_HOSTS | (empty) | Comma-separated production hosts to add to DNS rebinding allowlist |\n| MCP_ALLOWED_ORIGINS | (empty) | Comma-separated production origins to add to DNS rebinding allowlist |\n\nThe 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.\n\n## Tools reference\n\n### preview_nda_risk\n\nInput:\n- pdf_base64 (string): base64-encoded PDF bytes, max 10 MB\n- filename (string): the PDF filename\n\nOutput: 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.\n\n### get_nda_report\n\nInput:\n- session_token (string): from the preview_nda_risk response\n\nOutput: 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.\n\n## Disclaimer\n\nNDASentry 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.\n\n## Privacy\n\nThe hosted backend at ndasentry.ai is designed around an \"evaporates\" model:\n- Documents are processed in memory only, never written to disk\n- Reports are cached in RAM for 5 minutes after generation, then deleted\n- No account, no email collection, no document retention\n\n## About\n\nBuilt 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.\n\nIf you build something with NDASentry MCP, drop a note: **frontrangesupport@gmail.com**. Genuinely interested in what agents do with this.\n\n\n## License\n\nMIT — see [LICENSE](LICENSE) file in the repository root.\n",
  "bytes": 5328,
  "sha": "212a4e3074bfd9682a78fbc473e3fc6e7fd6b611206a67392139d6cfa6675afa",
  "repo_slug": "valtirman/ndasentry-mcp",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_valtirman_ndasentry_ce172ce5/readme"
}