Quick Verdict
DataNexus MCP is a hosted public-data intelligence service at https://datanexusmcp.com/mcp. Its Streamable HTTP endpoint gives AI agents tools for vulnerability and software-supply-chain research, nonprofits, domains and DNS, patents, US government contracts, regulatory records, NPI providers, FINRA brokers, and SAM exclusions. It connects public sources including NVD, CISA KEV, FIRST EPSS, OSV, IRS, USASpending, SAM.gov, NPPES, FINRA, EPO, USPTO, and WIPO. Provider claims such as “verified public data” do not guarantee that every record is current, complete, correctly matched, or sufficient for a decision.
As of July 21, 2026, repository and MCP metadata identify v2.4.12 and 55 tools. Three boundaries are critical. The source uses the Business Source License 1.1, which is source-available but not open source. The npm package depends on mcp-remote, so its local stdio process is a proxy to the hosted endpoint rather than the complete data engine running locally. Finally, official pages conflict on quotas, paid prices, and API-key headers. A production budget or authentication implementation cannot safely rely on one page without testing the live service.
Best For
- Security teams triaging CVEs, KEV status, EPSS, manifests, SBOMs, dependency graphs, license policies, and package-maintainer changes.
- Procurement, compliance, legal, and nonprofit researchers who need fast public-record screening inside an MCP client.
- Developers who preserve source links, query time, parameters, status, and human review rather than accepting the model narrative alone.
- Organizations prepared to assess BSL rights, remote processing, credential management, 90-day request logging, upstream availability, and vendor continuity.
- Not appropriate as an unattended engine for hiring, eligibility, diagnosis, treatment, legal opinions, sanctions, investment, or supplier rejection.
Key Features
- Vulnerability and supply-chain intelligence: NVD, CISA KEV, EPSS, OSV, deps.dev, npm, and PyPI support CVE summaries, SBOM audits, dependency checks, license analysis, maintainer history, and typosquatting detection.
- Nonprofit research: IRS and ProPublica filings, financial trends, executive compensation, and selected UK charity records. A generated health score is not an audit opinion.
- Domain and DNS: RDAP, DNS records, certificate transparency, subdomains, reverse IP, and SPF, DKIM, and DMARC checks.
- Patent and legal records: EPO, USPTO, and WIPO search by number, keyword, inventor, and citation. Patent validity and claim scope still need jurisdiction-specific professional review.
- Government and regulatory data: USASpending, SAM.gov, Regulations.gov, and Federal Register tools cover awards, solicitations, exclusions, dockets, and notices.
- Identity and compliance checks: NPI and FINRA lookups can accelerate screening, but names, entities, stale records, and missing disclosures require confirmation in the original registry.
- Structured agent output: tools return JSON or agent-ready Markdown and may report
OK,DEGRADED,NOT_FOUND, or validation errors. A partial response must remain visibly partial.
Use Cases
A security workflow can submit an SBOM or package manifest, screen for CVEs, KEV, EPSS, license conflicts, and maintainer anomalies, and then hand the evidence to an engineer. The reviewer must confirm the actual dependency path, exploit preconditions, patch, runtime exposure, and business impact. A SHIP/CAUTION/BLOCK label is a heuristic summary, not a release authority.
Procurement can combine US contract history with a SAM exclusion check, but entity identifiers, aliases, subsidiaries, query date, and jurisdiction must be verified. A grant reviewer can use IRS filings to locate financial trends and executive pay, then open the underlying filing and amendments rather than adopting a generated health score. Patent searches should preserve the query, family, claims, dates, citations, and legal status for counsel.
Expose only the necessary MCP tools and treat every remote field as untrusted input. Keep keys in client secret storage or environment variables, not prompts, repositories, or logs. Separate retrieval, normalization, model interpretation, expert review, and final approval. Never let one tool response automatically block a supplier, reject a person, delay healthcare, or publish a legal conclusion.
Pricing
The official information conflicts as of the review date. The homepage says 10 anonymous lookups per week and 100 per week with a free key. The documentation rate table says Free 500/month, Pro 5,000/month for $29, Team 25,000/month for $99, and Enterprise unlimited for $499. The repository README says 100 anonymous calls/month and 500 registered calls/month. Authentication examples also disagree: the docs use X-Api-Key, while the README uses X-DataNexus-Key. Production users should test no-key, invalid-key, rotation, revocation, exhausted-quota, and usage-counter behavior and obtain written commercial terms. These values should not be silently combined into one plan table.
The root LICENSE is BSL 1.1. Its Additional Use Grant allows personal projects, research, development, and internal self-hosting, but forbids offering the software or a substantially derived hosted, managed, or as-a-service product to third parties without a commercial license. The Change Date is June 11, 2030, with Apache-2.0 as the Change License. BSL is not an OSI open-source license. The repository’s pyproject.toml still declares MIT and an OSI classifier, which is an internal official-source conflict; rely on the root license and legal review rather than Python metadata.
Privacy is part of the price. The current policy says the hosted service records IP address, tool parameters, time, tool name, latency, outcome, and a random session identifier; API-key signup also collects email. Tool-call logs are retained for 90 days, while aggregated anonymized statistics may be retained indefinitely. Infrastructure is described as a private PostgreSQL database on Hetzner in the EU, with PostHog and Cloudflare also listed. Inputs can reveal investigation targets even when the underlying record is public. Minimize parameters and obtain written answers on backups, legal holds, subprocessors, deletion, and incident notice before sensitive use.
Pros
- One remote MCP endpoint screens several useful public-data domains without separate client integrations.
- CVE, KEV, EPSS, OSV, SBOM, license, and package-risk tools reduce repetitive security lookups.
- Named tools, source categories, and explicit status values support evidence capture and degraded-result handling.
- Anonymous and free-key access allow a small non-sensitive proof of concept.
- Readable source and an explicit internal-use grant support technical and legal diligence.
Cons
- Official homepage, documentation, and README disagree on quotas, pricing, key headers, and the meaning of local npm usage.
- BSL 1.1 is not open source and restricts third-party hosted derivatives; Python MIT metadata conflicts with the root license.
- The npm package is an
mcp-remoteproxy, so local installation does not keep queries or logs on the device. - Hosted processing retains IP and tool parameters for 90 days, which can expose sensitive research intent.
- Public sources can be delayed, revised, rate-limited, incomplete, or mis-matched, while generated verdicts and scores can produce false confidence.
Alternatives
| Tool | Better for | Main strength | Main boundary |
|---|---|---|---|
| Congress.gov MCP Server | Researchers focused only on US bills, members, and legislative activity | Narrower domain with a clear primary source | No equivalent vulnerability, domain, contract, or nonprofit catalog |
| Blockscout MCP Server | Web3 teams querying EVM addresses, transactions, and contracts | Deeper explorer and verified-contract tools | Not a general compliance service; its license also requires review |
| cyanheads OpenAlex MCP Server | Research on papers, authors, institutions, and citations | Focused scholarly knowledge graph | Not for vendor exclusions, CVEs, patent rights, or identity decisions |
| ClinicalTrials.gov MCP Server | Clinical trial registration and status research | Specialized study fields and registry context | Registry data is not clinical advice and has a much narrower domain |
| PubMed MCP Server | Biomedical literature search and summarization | Direct literature and biomedical-topic workflow | Literature does not establish patient-level evidence or compliance status |
FAQ
Is DataNexus MCP open source?
No. The root license is Business Source License 1.1, a source-available license rather than an OSI open-source license. It changes to Apache-2.0 on June 11, 2030; third-party hosted services are restricted before then.
Does running the npm package keep data local?
No. The current npm package depends on mcp-remote and bridges a local stdio client to the hosted MCP endpoint. A true self-host of the Python service requires separate deployment, upstream, license, and operational review.
What are the actual free and paid limits?
Official pages disagree, so this review cannot establish one reliable table. Check the live usage field and account, obtain a written quote, and test the quota period and over-limit response before production.
How long are queries logged?
The current privacy policy says IP, tool name, parameters, time, session, latency, and outcome are logged for 90 days. Aggregated anonymized statistics may remain indefinitely, and key registration adds email.
Can the output automate compliance or security decisions?
It should not. Identity matching, source freshness, upstream degradation, and risk algorithms can fail. High-impact decisions need original records, corroboration, qualified reviewers, and an appealable human process.
Bottom Line
DataNexus MCP is useful as an agent-accessible screening layer across several public databases, not as an automated truth service. Start with non-sensitive samples and only the necessary tools. Preserve sources and DEGRADED states, and test the current key header, quota counter, and exhaustion behavior. Then review BSL 1.1, distinguish the npm remote proxy from a full self-host, and decide whether 90-day hosted logs fit the use case. Security, procurement, medical, legal, financial, and personnel decisions must retain primary evidence, expert review, and accountable human approval.