Use the desk from your AI assistant.

Connect Claude, ChatGPT or your own AI agent software through MCP (Model Context Protocol, the open standard for linking AI tools to outside sources). It can then read the notes and search the official US record they cite: the screening lists, OFAC's recent actions, CISA's exploited-vulnerability catalog, the Federal Register and State travel advisories, with the New York Fed's reference rates and Treasury's par yield curve for the day a dated action has to be priced against. The cyber tools add CISA's advisories, incidents companies report to the SEC, new advisories against the software AI agents run on, a check of a codebase's own dependency versions against the vulnerability record, and what is new across GitHub, NVD and CISA's exploited-vulnerability catalog.

MCP

Read-only · no key

Endpoint

https://www.distributedrisk.com/mcp

The desk

list_notes
List desk notes
search_notes
Search desk notes
get_note
Read a desk note

The official record

search_screening_lists
Search the US screening lists
list_ofac_actions
OFAC recent actions
search_kev
Search CISA's KEV catalog
search_federal_register
Search the Federal Register
list_travel_advisories
State travel advisories
get_reference_rates
New York Fed reference rates
get_treasury_yields
Treasury par yield curve

Is this MCP server right for your work?

Distributed Risk provides a read-only MCP server for AI assistants researching sanctions, travel and cyber records. It gives the assistant the published desk notes, their sources and tools to search the underlying public record. It suits research that needs an inspectable source before a person makes a decision.

Where it fits

Use it for a deal team checking a counterparty, a credit committee following a company disclosure, a cyber underwriter examining an exploited vulnerability or a platform owner reading software advisories. The same source-backed notes are available on the public desk.

The server needs no key. Connecting it gives your assistant research tools; it does not give the assistant authority to approve a transaction, determine compliance or make a security finding about systems it has not examined.

Where the published tools fit
Research question	Published tools to inspect	Limit to keep in the answer
Does a counterparty name appear in US screening records?	`search_screening_lists`, `list_ofac_actions`	A name match requires identity checks; no match is not clearance
What is recorded about a vulnerability?	`get_cve`, `search_kev`	An exploitation record and an EPSS model estimate are different evidence
What has a public company disclosed about a cyber incident?	`list_cyber_incident_filings`	Read the filing; an optional disclosure is not automatically a materiality judgment
Do resolved dependency versions have known advisories?	`check_dependencies`	No advisory hit is not proof that a version is safe
What supports a desk assessment?	`search_notes`, `get_note`	Open the cited record and preserve its date

Public examples of the workflow

For a counterparty question, ask the assistant to show the matching list, identifiers, source link and the time the source was read, then compare those fields with the entity you mean. Escalate uncertainty to the person responsible for the decision.

For dependency research, the published tool accepts resolved package versions or a supported lockfile or SBOM. It compares them with the advisory record and returns the available severity, affected identifiers and fixes. A file that contains version ranges alone is not a resolved dependency list.

These are examples of how the published tools are intended to be used. They are not client results or a claim that your environment has been scanned.

What to compare before connecting

Check whether the server covers the source you need, whether the answer keeps a source link and date, and whether it distinguishes an official record from a model estimate or open-source advisory. Ask what happens when the source is unavailable rather than assuming an empty answer means nothing happened.

Keep private incident material and information your policy does not permit sharing out of any external tool request. Start from the specific public-record question you need answered.

Does this replace a compliance system or incident-response firm? No. The public tools support research. For work requiring an investigation, people on the ground or insurance, brief an advisor.

Where do I connect? Use the endpoint and client instructions in the Connect section below. Keep the official source links with any answer you use.

Decision support from official public sources. Not a legal determination of sanctions, travel, or cyber status. Readers own their compliance obligations.

Connect

Add a remote MCP server with this address. It speaks Streamable HTTP on protocol 2026-07-28, and negotiates the 2025 revisions with a client that opens with initialize. It needs no key.

Claude Code

claude mcp add --transport http distributed-risk https://www.distributedrisk.com/mcp

Clients that read an mcpServers file

{
  "mcpServers": {
    "distributed-risk": {
      "type": "http",
      "url": "https://www.distributedrisk.com/mcp"
    }
  }
}

From a terminal

Lists the tools the endpoint serves right now, which is the list below.

curl -s https://www.distributedrisk.com/mcp \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

The desk is listed in the official MCP registry as com.distributedrisk/desk, so a client that installs from the registry can find it there instead.

Tools

The desk

list_notes
Published desk notes, newest first: entry number, date (the day of the official action), published (the day the desk published the note), title, status, topic (cyber or sanctions), readers, source class, a one-line summary, and the note's URL. Pages with limit and offset; total is the count that matches, next_offset the offset of the next page. Filter by topic or reader as on the desk page.
search_notes
Find published notes that mention every word of the query (a name, a list, a country, a program, a date) in the title, summary, sources or body.
get_note
One published note in full (Markdown), with its numbered sources and links to the official record.

The official record

search_screening_lists
Search the Consolidated Screening List (OFAC SDN and other Treasury lists; BIS Entity, Denied Persons, Unverified and Military End User lists; State ITAR-debarred and nonproliferation lists) by name, alias or identifier (IMO, SWIFT/BIC, registration number). Close spellings match. Each result gives the list, programs, addresses, IDs, remarks and the official list link; the answer says when each list last changed.
Source: Consolidated Screening List, International Trade Administration
list_ofac_actions
OFAC's recent actions, newest first: designations and removals, general licenses, FAQs, regulations and settlements, each with its date, categories and the official page. Filter by date or words in the title.
Source: OFAC Recent Actions, Treasury
search_kev
CISA's Known Exploited Vulnerabilities catalog, newest first: CVE, vendor, product, date added, federal due date, known ransomware use, the required action, FIRST's EPSS estimate, and whether it sits in the AI and agent software stack. Look up one CVE, search vendor or product words, list what was added since a date, or only the AI stack. For one CVE in full, use get_cve.
Source: Known Exploited Vulnerabilities catalog, CISA
search_federal_register
Search Federal Register rules, proposed rules, notices and presidential documents: title, type, agencies, publication and effective dates, citation, abstract and the official document link. Narrow by agency (for example foreign-assets-control-office, industry-and-security-bureau, state-department), type or date.
Source: Federal Register API, Office of the Federal Register
list_travel_advisories
US State Department travel advisories, most recently updated first: country, level (1 to 4), the advisory's summary and the official page. Filter by country, minimum level or update date.
Source: Travel advisories, State Department
get_reference_rates
The New York Fed's reference rates: EFFR and OBFR (unsecured), SOFR, BGCR and TGCR (secured), and the SOFR averages and index (SOFRAI). With no rate, the latest of each; with a rate, its last N days, with percentiles, volume and, for EFFR, the target range.
Source: Reference rates, Federal Reserve Bank of New York
get_treasury_yields
Treasury's daily par yield curve, newest first: yields by tenor from 1 month to 30 years. Ask for the last N days, and optionally only some tenors (for example ["2 Yr", "10 Yr"]).
Source: Daily par yield curve rates, US Treasury

Cyber

CVE.org, CISA, NVD and SEC EDGAR are official. FIRST's EPSS is a model's estimate of exploitation, and OSV.dev and the GitHub Advisory Database are the open-source advisory record. Answers label each. A dependency version no advisory covers is not a version shown to be safe.

get_cve
Everything the record says about one CVE, in one answer: the CVE.org record (description, severity, weakness, affected products, references), CISA's exploitation call (SSVC) and its KEV listing with the federal due date, FIRST's EPSS estimate, and whether it sits in the AI and agent software stack.
Source: CVE.org (CVE Program), with CISA KEV and SSVC, and FIRST EPSS
list_cyber_advisories
CISA's advisories, newest first: alerts, joint cybersecurity advisories (#StopRansomware, state-sponsored activity), ICS and ICS-medical advisories, and analysis reports. Each with its kind, date, summary, the CVEs it names and the official page. Filter by kind, date, words, or the AI stack.
Source: Cybersecurity advisories, CISA
list_cyber_incident_filings
Cybersecurity incidents public companies report on Form 8-K, newest first: Item 1.05 (the company judged it material) and, unless excluded, Item 8.01 reports made without that judgment. Each with the company, tickers, filing and event dates, items, state, industry code and the EDGAR filing.
Source: EDGAR full-text search, SEC
list_ai_stack_vulnerabilities
New advisories against the software AI agents run on, newest and most severe first: agent frameworks (LangChain, LlamaIndex, CrewAI, n8n, Langflow), LLM gateways (LiteLLM), MCP servers and SDKs, model serving (vLLM, Ollama), ML platforms (MLflow, Ray, Jupyter), model libraries, AI apps and coding agents. Malicious-package reports (a compromised or lookalike release) come first. CISA KEV entries in the same stack are included. Each with package, severity, fixed versions, CVEs, EPSS and the advisory.
Source: OSV.dev (GitHub, PyPI and OpenSSF advisories), with CISA KEV
check_dependencies
Which of these package versions have a known vulnerability. Give packages directly (ecosystem, name, version), or paste a lockfile or SBOM as manifest and it is read for you: package-lock.json, pnpm-lock.yaml, yarn.lock, requirements.txt, poetry.lock, Pipfile.lock, go.mod, go.sum, Cargo.lock, Gemfile.lock, composer.lock, packages.lock.json, gradle.lockfile, cyclonedx, spdx, purls. Each version is checked against OSV.dev, which aggregates GitHub's reviewed advisories, the Go and PyPI databases, RustSec and OpenSSF's malicious-package reports across npm, PyPI, Go, Maven, crates.io, NuGet, RubyGems, Packagist and more. Every hit comes back with its severity, the CVEs it is known by, the versions that fix it, whether CISA lists it as exploited with the federal due date, and FIRST's EPSS estimate. Exploited packages come first, then malicious packages, then severity. A version with no hit is a version no advisory in those databases covers; that is not a statement that it is safe. Only resolved versions are read: a file of ranges (package.json, Pipfile, pom.xml) cannot be checked, so pass the lockfile beside it.
Source: OSV.dev (GitHub, Go, PyPI, RustSec and OpenSSF advisories), with CISA KEV and FIRST EPSS
list_new_vulnerabilities
What is new across the three vulnerability feeds, exploited first, then by severity, then newest: GitHub's reviewed advisories (the affected package range and the first patched version), NVD's newly published CVE records (software no package ecosystem carries, with NIST's scoring), and the CVEs CISA added to its Known Exploited Vulnerabilities catalog. A KEV addition keeps its own date: it is CISA saying this one is being exploited now, often long after the CVE was published. A CVE that GitHub and NVD both published in the window appears once, from GitHub. For CISA's alerts and joint advisories rather than vulnerabilities, use list_cyber_advisories.
Source: GitHub Advisory Database, NVD (NIST) and CISA KEV

What it returns

Two kinds of answer. From the desk: each note's entry number, date, status, sources and full text, with a link back to its page. From the official record: what the government source publishes, as it publishes it, with the source link and the time we read it. A screening-list answer also says when each list last changed. A CVE answer joins the CVE.org record, CISA's KEV listing and exploitation call, and the EPSS estimate, each dated. A rates or yield-curve answer gives the business day each figure is for, which is not always the day you asked.

A list match is a lead to check against the official entry, not a determination that a party is listed. No match is not a clearance: names vary in spelling and transliteration, and the lists change.

The desk's terms apply to every answer, and every tool result carries them: decision support from official public sources, not a legal determination of sanctions, travel, or cyber status. Readers own their compliance obligations.

Every tool only reads, and none of them can see requests sent to an advisor. One address can make about 120 calls a minute, and a request body can run to 64 KB. A newly published note reaches connected assistants within a minute. The lists and catalogs are read again every hour, OFAC's recent actions every 20 minutes, CISA's advisories, SEC filings and the reference rates every 30 minutes, and the Federal Register every 10. The first list search after a quiet spell can take a few seconds while the lists load.

Beyond the record

Some decisions need people on the ground, an investigation, or insurance. Tell us what you need and where, and an advisor arranges it with a partner firm that works in that country.

Brief an advisor