@cyanheads/sanctions-screening-mcp-server
Screen names against eleven OFAC, EU, UK, UN, BIS, and State sanctions and export-control lists and resolve legal entities against GLEIF, matched offline over a local SQLite + FTS5 mirror. A screening aid, not a compliance determination.
7 Tools • 4 Resources • 1 Prompt
Overview
Sanctions and export-control screening over eleven lists plus the GLEIF LEI registry, matched offline against a local mirror. Screen a name or identifier for potential watchlist hits, resolve a company to its LEI, and trace its accounting-consolidation relationships.
The six BIS/State lists come from trade.gov's keyless Consolidated Screening List (CSL) bulk download. Treasury CSL copies are skipped; direct OFAC provenance and grouping remain unchanged.
[!IMPORTANT]
A screening aid, not a compliance determination. Every result is a potential match with a transparent score and source provenance, never a verdict. A hit is a candidate to verify against the official source; an empty result is never a clearance. Sanctions compliance needs human review and a qualified determination. This server feeds that process and does not perform it.
Tools
Resources
The three sanctions:// data resources are also reachable through the tools, for clients that don't surface resources. The guide is also published in the plugin, npm package, and bundle under skills/sanctions-screening-guide/.
Prompts
Capability reference
sanctions_screen_name tool
name in any script (at most 64 words and 1,024 characters) plus optional sources, entityType, and minScore filters; matchMode is strict (default) or fuzzy. Strict runs a fuzzy pass over each selected list it finds nothing on, and over the UK and UN lists either way for a name holding a letter outside the Latin script: a full pass when it finds nothing on any list, otherwise one that adds, after the strict hits, only candidates covering every word of the name, by spelling or by sound, other than legal forms, articles and other function words, and the jurisdiction codes uk, usa, uae, and rf. fuzzySources names the lists a fuzzy pass searched; up to 100 hits per page (limit, default 25) with offset
- Limits above 100 clamp with a notice;
SANCTIONS_NORMALIZE_INPUTS=false restores rejection above the maximum
- Hits carry
source, sourceEntryId, the list's published referenceNumber where it has one, unReferenceNumber on a UK or EU hit whose list publishes one, matchedName with its matchedNameType, and matchType (exact / strong / approximate); strong is every word of the name present, a legal form counting in any spelling of the same form (Ltd and Limited, LLC and OOO; never LLC and Ltd)
- Approximate hits add the raw Jaro-Winkler
score (0–1) and queryTokenCoverage: covered counts the words matched by spelling or by sound, and phonetic (absent when zero) how many of those matched by sound alone, through an equal primary Double-Metaphone key of three or more characters (gaddafi and qadhafi), neither word a legal form, article or other function word, or jurisdiction code. A sound match never admits a candidate or raises a score, and of two hits with equal score and coverage the one with fewer sound matches ranks first
- A typed screen includes records whose source omitted the entity type; those hits remain
unknown. Published birth dates (with circa where supplied), nationalities, and distinct address countries appear on each hit when available, preserving date precision and ranges for comparison before designation lookup
- Approximate screening compares native forms and deterministic Russian-alphabet romanizations in both directions, always keeping the name and alias type exactly as published. Derived equality remains approximate; it is neither an official alias nor an identity determination. Native candidates take priority within each list's independent 2,000-name fuzzy pool; romanized forms and first-four-code-point deletion signatures supplement remaining capacity, so typo recall is bounded
sources names every selected list a hit's party is on. OFAC publishes a party on the SDN List and a non-SDN list under one entry ID in both of its files; with both OFAC lists selected that party is one hit, counted once, naming both even when the screen matched one record. Its other fields, the designation date included, come from the source record: ofac_sdn, unless the Consolidated record matched alone or better
totalAvailable, hasMore, and nextOffset page the rest, and paging reaches every match counted; totalAvailableBasis marks the count exact (no fuzzy pass ran) or a lower_bound (a fuzzy pass ran), and a notice asks for a more distinctive word when a list's fuzzy candidate pool reached its budget
sanctions_screen_identifier tool
value (an identifier as you hold it) plus optional type (any default, imo, swift_bic, digital_currency_address, passport, national_id) and sources
- Exact match after normalization, never fuzzy and never scored: spacing, letter case, and
- . / are ignored; an IMO number matches with or without its IMO prefix; a SWIFT/BIC compares on its first eight characters, so a branch BIC11 matches its institution's BIC8; a wallet address folds case only for hex (0x…), bech32 (bc1…, ltc1…, bnb1…), and cashaddr encodings, while base58 addresses compare exactly
type matches on each list's published label: imo includes OFAC Vessel Registration Identification, UK IMO Number, and EU IMO (vessel identification); an otherwise unclassified label, including IMO Company Number, also qualifies when its normalized value is IMO followed by seven digits. Established passport, national-ID, SWIFT/BIC, and wallet labels retain their categories. swift_bic is SWIFT/BIC and SWIFT BIC; digital_currency_address is OFAC Digital Currency Address - <code>; passport and national_id cover the lists' passport and national-ID labels. any also reaches every label with no category (MMSI, call signs, tail numbers, tax and registration numbers, email, websites)
- One hit per designation, ordered by list then entry ID, each with the
matchedIdentifiers that matched as published and its sources, an OFAC party on both OFAC lists being one hit naming both; not paged. An identifier a list prints only in free-text remarks, or bundled with other numbers in one field, does not match
sanctions_get_designation tool
source (one of the eleven codes above) and entryId: the sourceEntryId from a screening hit, or the reference number the list publishes (UN QDe.004, EU EU.27.28, UK OFSI Group ID 14196) or, on uk and eu, the UN reference number the entry implements (TAe.010). Matched trimmed and case-insensitive: entry ID first, then reference number, then UN reference number, then a UN reference number in another zero padding (LYi.28, LYi.028), each only when the earlier ones match nothing. A reference number or UN reference number two designations share fails as reference_ambiguous, naming both. CSL records retain the provider's opaque id; missing entity type remains unknown
- Returns
referenceNumber beside sourceEntryId where the list publishes one; OFAC publishes none, its entry ID being its published number
- Returns
unReferenceNumber for a UK or EU designation whose list publishes one, as published and never corrected: it can differ from the UN list's own in case or padding (a lookup ignores both), name no current UN listing, or be carried by two designations. It is a link to check against the un record, not proof of identity
- The EU list publishes no primary name, so
primaryName is its first Latin-script alias carrying a function; with none, when the first alias is in another script, the first Latin-script EN alias, then the first untagged Latin-script alias, then the first Latin-script alias. Every other published name follows as an alias in document order
- All published aliases, including the UK list's non-Latin-script names (typed
aka) and the UN list's original-script names (typed primary), identifiers, addresses, dates and places of birth, nationalities, descriptive features, program, legal basis, and designation date; a field the source omitted is absent, never filled in
- Identifiers cover identity documents plus the SWIFT/BIC codes, digital-currency addresses, vessel call signs, aircraft tail and serial numbers, phone numbers, email addresses, and websites a list publishes, each typed with the list's own label and its value verbatim
- Dates of birth are ISO 8601 at the precision the source published (
1952-10-07, 1946-08, 1938, or an interval such as 1955/1957), with circa: true where the source marks one approximate; designationDate normalizes valid complete dates to YYYY-MM-DD, retaining unparseable published text. CSL uses its published start_date, without substituting dates from remarks
- An OFAC party is dated from its entries on the lists its file publishes: the SDN List for
ofac_sdn, the non-SDN lists for ofac_consolidated, earliest first. A party on both files can carry a different date on each
legalBasis is the instrument the list cites, verbatim: OFAC's legal-basis references (Executive Order 14024 (Russia)), the EU regulation title (2024/1488 (OJ L27052024)), or the UK regime's regulations (The Russia (Sanctions) (EU Exit) Regulations 2019), several joined with ; . The UN list publishes none
features carries what a list publishes to describe the party rather than identify it, each typed with the list's own label, in published order: OFAC's vessel flag, type, owner, and tonnage, aircraft model, operator, and manufacture date, title, gender, and sanctions notes (Secondary sanctions risk: See Section 11 of Executive Order 14024.), and the UK list's designation source and measures imposed (DesignationSource, SanctionsImposed, one feature per |-separated part), then its ship details (CurrentBelievedFlagOfShip, TypeOfShip, YearBuilt, …). A date is ISO 8601 at published precision, with circa: true where OFAC marks it approximate. The EU and UN lists contribute none, and no screening tool matches a feature value. CSL features also retain published restrictions, order flags, end dates, country, notices, and provenance URLs without legal interpretation.
sanctions_resolve_entity tool
name in any script (same bound as sanctions_screen_name) plus optional jurisdiction, status, and minScore; matchMode is strict (default), which falls back to a fuzzy pass only when strict finds nothing, or fuzzy; up to 50 candidates per page (limit, default 10) with offset
- Limits above 50 clamp with a notice;
SANCTIONS_NORMALIZE_INPUTS=false restores rejection above the maximum
jurisdiction is a country code, which matches the country and every subdivision under it (US matches US-DE and US-CA), or an ISO 3166-2 subdivision code (US-DE), matched exactly; case-insensitive
status: issued (default) matches ISSUED, lapsed matches exactly LAPSED, and any applies no filter — the only way to reach RETIRED, DUPLICATE, ANNULLED, PENDING_TRANSFER, PENDING_ARCHIVAL, or MERGED records, each candidate's status naming its state
- When a status filter excludes every strict match, a notice counts distinct excluded LEIs by registration status and points to
status: "any". It retains the jurisdiction filter, marks capped counts as lower bounds, and leaves fuzzy fallback and ranking unchanged
- Searches every name GLEIF publishes: the legal name, previous legal names, trading names, alternative-language legal names, and ASCII transliterations of a legal name in another script
- Candidates carry
lei, legalName, the matchedName and its matchedNameType (LEGAL_NAME, PREVIOUS_LEGAL_NAME, TRADING_OR_OPERATING_NAME, …, or UNKNOWN for a name stored without a type), and matchType, with score and queryTokenCoverage on approximate matches; one candidate per LEI, paged by the same totalAvailable / totalAvailableBasis / hasMore / nextOffset fields
sanctions_get_entity tool
- One 20-character
lei; returns the legal name, otherNames, legal and headquarters addresses, status, jurisdiction, registration authority, and lastUpdate
entityStatus is GLEIF's EntityStatus as published (ACTIVE, INACTIVE, NULL), the legal entity's own state where status is the LEI's. successors lists every successor entity GLEIF names, each by lei, name, or both: a retired, merged, or duplicate LEI can name the entities that took it over. Successors are listed, never screened or followed; pass a successor's LEI to sanctions_get_entity or sanctions_trace_ownership. entityStatusLoaded is false for a record stored before status and successors were read, which says nothing about whether it has either; a GLEIF refresh that updates the record, or mirror:init, loads them
- Mixed-case LEIs normalize to uppercase for this tool, ownership tracing, and the entity resource, including cursor binding.
SANCTIONS_NORMALIZE_INPUTS=false restores uppercase-only validation. Check digits are checked after a lookup miss, so held records remain accessible
alternateNames lists every other and transliterated name with its GLEIF type; a name the mirror stored before types were kept reads as UNKNOWN
sanctionsHits cross-references the entity against every watchlist: the legal name and every alternateNames entry screened strict (exact, then all tokens present with a legal form matching any spelling of the same form, never fuzzy), and the LEI and registrationAuthorityEntityId looked up exactly as non-document identifiers (never passports, national IDs, IMO numbers, SWIFT/BIC codes, or wallets). The registration number matches only an identifier published for the country of the entity's jurisdiction, and is skipped when the entity publishes no jurisdiction or a not-available placeholder (N/A, n.a.) in place of a number
- One hit per designation, an OFAC party on both OFAC lists being one hit whose
sources names both, exact name and identifier matches first, then strong name matches, capped at 25 after the merge. matchedOn lists every input that produced a hit (legal_name, other_name with its GLEIF type, lei, registration_number); an identifier match adds matchedIdentifiers as the list publishes them, and a hit no name produced carries no matchedName or matchType. No hit carries a score: only an approximate (fuzzy) match has one, and every screen here is strict
screeningStatus (screened / not_ready) says whether the cross-reference ran; sanctionsScreen counts distinct designations across every input (totalAvailable), flags a capped list (hasMore), and lists in screenedInputs what was screened beyond the always-screened legal name and LEI
sanctions_trace_ownership tool
- Root
lei, direction (parents / children / both, default both), and depth 1–5 (default 3); screenNodes: true cross-references every node as sanctions_get_entity does (its names strict, its LEI and country-matched registration number as identifiers), up to 10 merged hits per node; a node with no Level 1 record has no name to screen and is looked up by its LEI alone. both is a parents walk plus a children walk from the root, each to depth; it never reaches siblings or co-parents
limit defaults to 50 and clamps above 100 records across nodes, then edges, then boundaries; every page but the last holds exactly limit records. Drain nextCursor while hasMore is true; totalAvailable counts those segment records exactly. Edges reference nodes in the assembled segment across pages. Only returned nodes are hydrated and screened. Page size follows the record count alone, so lower limit for smaller pages: a screened node at its ten-hit cap is the largest record
- Cursors bind the effective query and topology fingerprint; invalid or stale cursors, and a cursor carrying a key the server did not write, require restarting the segment. Entity and screening values remain current between pages, without an immutable snapshot
role and depth retain segment placement; ultimate-parent shortcuts are unwalked leaves flagged reachedVia: "ultimate". After draining a segment, re-root each boundary with its supplied direction and depth; deduplicate queued (lei, direction) roots and edge identities because segments overlap. Depth restarts per segment; a missing Level 1 root is accepted only if Level 2 references it
truncationReasons distinguishes remaining pages, depth limits, and ultimate leaves. complete requires no remaining page or omitted branch and no missing Level 1 record anywhere in the segment (missingEntityCount); missingEntityLeis is page-local. This covers loaded accounting-consolidation relationships only, never beneficial ownership
- Each node whose parents the walk read carries
parentStatus.direct / .ultimate: relationship, exception (a GLEIF reporting exception, with every reason, such as NATURAL_PERSONS), none, or unknown. unknown means exception data is not loaded, which reportingExceptionsLoaded: false states.
screeningStatus (screened / not_requested / not_ready), screenedNodeCount, and flaggedNodeCount; each screened node carries its own sanctionsScreen (hasMore, screenedInputs) and hits with matchedOn and matchedIdentifiers, in the sanctions_get_entity shape
sanctions_list_sources tool
- No input; one row per sanctions list plus
gleif, each with recordCount, url (the configured source URL), and license; the gleif row's recordCount is Level 1 entities and its relationshipCount Level 2 ownership relationships
sanctionsReady / sanctionsAsOf and leiReady / leiAsOf report whether each mirror has synced and when; sanctionsAsOf is the last successful local sync in which every sanctions list refreshed, never upstream publication freshness. Not gated on readiness, so it reports an empty mirror instead of failing
- Each sanctions row carries
lastHarvestedAt, when this server last harvested the list successfully, and publishedAt, the date the publisher put on that document at the precision and offset published (OFAC DateOfIssue, EU generationDate, UK DateGenerated, UN dateGenerated, each CSL list's source_last_updated). publishedAt dates the document, not the last change to the list: the UN file is regenerated daily. Both are absent until this server records a harvest of the list, which means unknown, not stale; publishedAt is also absent when the document carried no date this server reads. GLEIF rows carry neither
lastRefresh appears only while a sanctions refresh is running (in_progress, which a killed refresh also reads as) or after one failed (error, with failedSources naming the lists that failed: empty when none failed and the run stopped for another reason, absent when which failed is unknown). A list that failed keeps its earlier lastHarvestedAt and publishedAt
reportingExceptionsLoaded, and on the gleif row a reportingExceptionCount once the reporting exceptions are loaded
alternateNamesIndexed: whether GLEIF's trading, previous, alternative-language, and transliterated names are indexed; until mirror:init builds the index, sanctions_resolve_entity searches legal names only
entityStatusLoaded: whether a completed mirror:init golden-copy load has written entity status and successors for every entity in that copy; a sanctions_get_entity record's own entityStatusLoaded is the per-record answer
sanctions://designation/{source}/{entryId} resource
source is one of the eleven list codes, entryId the list's own ID, its published reference number, or (UK and EU) the UN reference number the designation implements, resolved as sanctions_get_designation resolves it and decoded once when percent-encoded; returns the sanctions_get_designation payload as application/json
- Cached for an hour, scoped
private
sanctions://entity/{lei} resource
- The
sanctions_get_entity Level 1 payload for one lei, without the sanctions cross-reference, which is tool-only
- Cached for an hour, scoped
private
sanctions://sources resource
- The
sanctions_list_sources payload, unchanged
ttlMs: 0: never cached, because readiness and the as-of timestamps are the payload
sanctions_vet_counterparty prompt
- Arguments:
name required; jurisdiction optional (a country code, which includes its subdivisions, or an ISO 3166-2 subdivision code)
- Returns one user message that screens the name (and any identifier the caller holds, with
sanctions_screen_identifier), resolves it to an LEI (treating a retired, merged, or duplicate LEI's successor as the entity to vet), traces ownership with screenNodes: true, pulls each hit's designation record, and asks for a summary that treats every match as a candidate to verify and says, from matchedOn, which of the entity's names or identifiers produced it. It names GLEIF parents as accounting-consolidation parents, not beneficial owners, and reports a reporting exception's reasons rather than reading it as "no parent"
Features
Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.
Sanctions-screening-specific:
- One screen covers the eleven sanctions and export-control lists above; the matching list shows up as per-hit provenance
- Offline and keyless, with no per-request upstream rate limit: all lists are normalized into local SQLite + FTS5 mirrors via the framework
MirrorService
- Per-alias name index, so a query matches any of an entity's names in one FTS scan, and a per-identifier index for exact lookup by IMO number, SWIFT/BIC, wallet address, or document number
- Strict-then-fuzzy matching: exact-normalized, then all tokens present (FTS5, a legal form matching any spelling of the same form), then Jaro-Winkler + Double-Metaphone over a 2,000-candidate pool per list, so adding a list never pushes another list's candidates out, with every admitted candidate reachable by paging
- GLEIF Level 1 (who is who) and Level 2 (who owns whom) for entity resolution and ownership tracing
Agent-friendly output:
- Real signal with provenance: approximate hits carry the raw Jaro-Winkler
score and a literal queryTokenCoverage count, never a blended confidence; every hit names its list, program, designation date, and the name that matched, typed primary / aka / fka / low-quality-aka
- Decision-support
caveat in the output of sanctions_screen_name, sanctions_screen_identifier, sanctions_get_designation, sanctions_get_entity, and sanctions_trace_ownership
- Disclosed gaps:
totalAvailableBasis, screeningStatus, complete / truncated / missingEntityLeis, and each node's parentStatus say what a response did not cover
- Typed errors: every data tool and
sanctions:// resource except the sources listing fails as mirror_not_ready (retryable) until its mirror is loaded; unknown IDs fail as designation_not_found or lei_not_found, an LEI the mirror lacks whose ISO 17442 check digits fail as invalid_lei_checksum (a mistyped character), and a reference number or UN reference number two designations share as reference_ambiguous; a name with no letter or digit fails as name_not_searchable, one past the length bound as name_too_long, and an identifier with nothing left once spacing and separators are removed as identifier_not_searchable
Getting started
Public Hosted Instance
A public instance is available at https://sanctions-screening.caseyjhand.com/mcp — no installation required. Point any MCP client at it via Streamable HTTP:
Self-Hosted / Local
Add the following to your MCP client configuration file. The mirror is not bundled, and the npm package cannot build it: run bun run mirror:init from a clone (Installation) or inside the Docker image, then point SANCTIONS_MIRROR_PATH at the result (see Mirror lifecycle).
Or with npx (no Bun required):
For Streamable HTTP, set the transport and start the server:
Prerequisites
- Bun v1.4.4 or higher (or Node.js v24+).
- Disk for the local mirror; GLEIF Level 1 takes most of it. No source needs an API key.
Installation
- Clone the repository:
- Navigate into the directory:
- Install dependencies:
- Configure environment:
- Populate the mirror:
Configuration
Every source is keyless, so nothing here is required.
See .env.example for the full list of optional overrides.
Running the server
Local development
-
Build and run:
-
Run checks and tests:
Mirror lifecycle
Upgrading from 0.5.1 or earlier: run bun run mirror:refresh to load the six BIS/State CSL lists; optionally set SANCTIONS_REFRESH_SKIP_GLEIF=1. Existing readiness does not prove those lists are populated. sanctions_list_sources reports configured lists and stored counts, with refresh guidance for zero-row CSL sources.
The mirror loads out-of-band, never on the request path: bun run mirror:init does the full load once, bun run mirror:refresh keeps it current (HTTP deployments also refresh on SANCTIONS_REFRESH_CRON), and bun run mirror:verify reports readiness and counts. When upgrading from an earlier release, run mirror:refresh once (from 0.3.0 or earlier, mirror:init). Failure handling, upgrade and rollback behavior, and disk sizing are in docs/mirror.md.
Docker
The Dockerfile defaults to HTTP transport, stateless session mode, and logs to /var/log/sanctions-screening-mcp-server. Mount a volume at /usr/src/app/data so the mirror survives restarts, and populate it with bun run mirror:init via docker exec. OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.
Project structure
Development guide
See CLAUDE.md/AGENTS.md for development guidelines and architectural rules. The short version:
- Handlers throw, framework catches — no
try/catch in tool logic
- Use
ctx.log for request-scoped logging; the mirror is reached through getScreeningService(), not ctx.state
- Register new tools and resources via the barrels in
src/mcp-server/*/definitions/index.ts
- Wrap external sources: validate raw → normalize to the common schema → return the output schema; never fabricate fields a source omits, and never synthesize a confidence score
Attribution
This server redistributes open data from these sources, cited per their terms:
UKSL has been the single authoritative UK source since the OFSI Consolidated List closed on 28 January 2026.
This product uses the International Trade Administration’s Data API but is not endorsed or certified by the International Trade Administration.
Verify CSL candidates and their published restrictions against the official agency lists and Federal Register publications. A published end date or order flag is retained as source data, never interpreted as clearance or a legal-status determination. Mirror timestamps report successful local synchronization. A list's publishedAt is the publisher's date on the document harvested, not proof of when the list's content last changed.
Contributing
Issues are welcome. Run checks and tests before submitting:
License
Apache-2.0 — see LICENSE for details.