trust-check-mcp
Free wallet-OFAC SDN screening and trust-check preview for AI agents on Base. Single MCP tool call returns allow/block verdict against US Treasury OFAC SDN list.
README
trust-check-mcp
Open client and API spec for PaladinFi Trust Check — free wallet-OFAC SDN screening and sample-fixture trust-check preview for AI agents on Base. This repository contains the public REST and MCP API specification, working code examples, and thin client wrappers. The hosted backend is proprietary.
Operated by Malcontent Games LLC, doing business as PaladinFi.
What is PaladinFi Trust Check?
A free wallet-OFAC-SDN screening surface for AI agents on Base. Single MCP tool call returns a binary allow / block verdict against the US Treasury OFAC SDN list.
- No auth surface. No API key, no signup, no billing.
- Live US Treasury OFAC SDN data. Refreshed daily from the Treasury XML feed.
- Freshness metadata per response.
_ofac_list_updated_at(ISO timestamp) and_ofac_sdn_countexposed in every response (inside thetrustblock); callers can verify staleness without trusting our bookkeeping. - Signed responses. Every
/ofacresponse carries an Ed25519 signature (_signature,_signature_alg,_signature_pubkey_hex) over the canonical JSON body, so a consumer can verify origin integrity. - Rate limit. 1 request per second per IP + burst 3 + 3 concurrent connections per IP.
- Non-custodial, read-only. No money handling, no calldata, no signing surface, no transaction execution.
This MCP is the off-chain HTTP surface for sanctions screening. For in-contract gate logic (Solidity modifiers running inside settlement contracts), on-chain sanctions oracles remain the canonical surface.
Install (MCP)
For Claude Code or any MCP-compatible client supporting Streamable-HTTP transport:
claude mcp add --transport http --scope user paladin-trust https://trust.paladinfi.com/mcp
Verify the host is up before adding:
curl -i https://trust.paladinfi.com/health
# Expect: HTTP/1.1 200 OK
After mcp add, restart your client. Three tools become available:
trust_check_ofac_free(address, chainId?)— live US Treasury OFAC SDN wallet-screen. Returnsrecommendation: "allow" | "block"with_real: trueplus freshness metadata.trust_check_preview(address, chainId?)— sample-fixture preview of the broader trust-check response shape. Thetrustblock carries_preview: true, every factor isreal: false, and recommendation is prefixedsample-. Do not use the preview verdict to gate real swaps, signing, or any production agent decision.trust_check_health()— liveness, supported chains, OFAC list refresh timestamp + size.
See mcp-tools.json for the full tool schemas.
REST API
The MCP server fronts three REST endpoints. The backend is hosted alongside other PaladinFi services on a shared origin (swap.paladinfi.com); this MCP is scoped to read-only OFAC screening and does not expose calldata, signing, or transaction surfaces from the broader platform. The REST endpoints can be called directly via curl:
# Free OFAC SDN screen
curl -X POST https://swap.paladinfi.com/v1/trust-check/ofac \
-H "content-type: application/json" \
-d '{"chainId": 8453, "address": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913"}'
# Sample-fixture preview
curl -X POST https://swap.paladinfi.com/v1/trust-check/preview \
-H "content-type: application/json" \
-d '{"chainId": 8453, "address": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913"}'
See openapi.yaml for the full schemas.
chainIdnote: the OFAC SDN screen is wallet-address-level and chain-agnostic — the same 0x address is listed (or not) regardless of network.chainIdis accepted for consistency with the full/v1/trust-checkendpoint; it does not scope the OFAC match.
Coverage and limits
| Dimension | Value |
|---|---|
| Chain | Base mainnet (chainId 8453) |
| Source | US Treasury OFAC SDN XML feed |
| Jurisdiction | US OFAC SDN only. No EU, UN, UK, or other sanctions regimes; callers in non-US jurisdictions are responsible for screening against locally-applicable lists. |
| Refresh cadence | Daily from the Treasury XML feed |
| List size | ~93 wallet/EOA addresses (see live _ofac_sdn_count) |
| Rate limit | 1 req/sec + burst 3 + 3 concurrent per IP |
| Auth | Anonymous (no API key, no signup) |
| Cost | Free |
Response shape
trust_check_ofac_free
The freshness + scope markers live inside the trust block. The top level also carries an Ed25519 signature and a pointer to the paid full-composition endpoint.
{
"address": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
"chainId": 8453,
"trust": {
"recommendation": "allow",
"factors": [
{ "source": "ofac", "signal": "not_listed", "weight": 0, "details": "", "real": true }
],
"version": "1.1",
"_real": true,
"_scope": "ofac-only (wallet-address screen; use /v1/trust-check for full composition: GoPlus + Etherscan + anomaly heuristics)",
"_ofac_list_updated_at": "2026-05-31T04:08:06Z",
"_ofac_sdn_count": 93
},
"_signature": "<base64 Ed25519>",
"_signature_alg": "ed25519",
"_signature_pubkey_hex": "272b6b62230d9da810f3ed64b5e5147f1ae062cb46ad7f25044b0aab1d18fb6f"
}
A sanctioned address returns the same shape with trust.recommendation: "block" and a factor { "source": "ofac", "signal": "sdn_listed" }.
Programmatic safety check before consuming the verdict:
resp = httpx.post("https://swap.paladinfi.com/v1/trust-check/ofac",
json={"chainId": 8453, "address": addr}).json()
trust = resp["trust"]
assert trust["_real"] is True # not a fixture
assert trust["_scope"].startswith("ofac-only") # not a different evaluation path
verdict = trust["recommendation"] # "allow" or "block"
trust_check_preview
A SAMPLE FIXTURE — not a live evaluation. Same envelope (address, chainId, trust, plus top-level request_id and _mcp_paid_endpoint_info), but inside trust: _preview: true, recommendation is prefixed sample- (e.g. "sample-allow"), risk_score is null, every factor is real: false, and trust._message carries a do-not-use-as-real-verdict banner. There is no _real field on preview — the _preview: true marker and the sample- prefix are how you tell a fixture from a real screen.
Refresh cadence
The OFAC SDN list is fetched from https://www.treasury.gov/ofac/downloads/sdn.xml daily. The exact timestamp of the in-memory list is returned per-response in trust._ofac_list_updated_at, so callers can detect stale data without trusting a separate bookkeeping layer.
If the source feed is unreachable, the in-memory list is held over and _ofac_list_updated_at continues to reflect the last successful refresh. Stale lists do not auto-degrade to block — callers responsible for enforcing maximum-acceptable staleness check _ofac_list_updated_at themselves.
Logging and retention
The server records request timestamp, source IP (for rate-limit accounting only), and the queried address in standard nginx access logs. Logs are retained for 30 days for operational debugging and then rotated. No persistent association is built between caller identity and queried addresses beyond the rate-limit window. No analytics, no third-party tracking, no PII enrichment.
Rate-limit behavior
Requests exceeding 1 r/s + burst 3 receive HTTP 429. The rate-limit window is per-IP; there is no IP ban or escalating block — the next request after the window clears is served normally. The Retry-After header indicates when to retry.
Use of this endpoint is subject to the PaladinFi Terms.
Related packages
For evaluations that compose multiple signals (OFAC + GoPlus token security + Etherscan source verification + anomaly heuristics), PaladinFi offers other npm packages — see paladinfi.com/trust-check for details. Migration from this MCP to one of the multi-signal packages: install the chosen npm package alongside (or instead of) this MCP; tool names overlap so callers can migrate without rewriting call sites.
When to use this vs. an on-chain sanctions oracle
| On-chain oracle | trust_check_ofac_free |
|
|---|---|---|
| Access | Solidity call on a settlement contract | HTTP POST (anonymous) |
| Best for | In-contract gate logic (modifiers) | Off-chain agents, bots, web apps, MCP clients |
| Response | Bare bool |
Structured JSON with freshness metadata |
| Cost | Gas per call | Free + rate-limited |
| Coverage | Often multi-jurisdiction | US OFAC SDN only |
These are complementary surfaces: an off-chain HTTP screen for agent / web-UI display, and an on-chain modifier for the settlement contract.
Examples
Zero-dependency, copy-paste runnable. Each defaults to screening one clean address and one genuinely OFAC-listed address, so you see both an allow and a block on first run. Pass your own addresses as arguments. Exit code: 0 all clean, 1 any address listed, 2 request error — drops straight into CI or a pre-send gate.
examples/python/check_address.py— Python 3 (standard library only; nopip install). Run:python check_address.py.examples/typescript/check_address.ts— TypeScript with built-infetch. Run:npx tsx check_address.ts.
For a browser/React integration of the same endpoint, see the tutorial at paladinfi.com/docs/screen-wallets/.
Documentation
- Hosted MCP endpoint: trust.paladinfi.com/mcp
- Live
/health: trust.paladinfi.com/health - Apex docs: paladinfi.com/trust-check/
- Sister MCP for swap routing: paladin-swap-mcp
Contact
- General: dev@paladinfi.com
- Security: see SECURITY.md
Legal
Operated by Malcontent Games LLC, doing business as PaladinFi. The hosted backend at swap.paladinfi.com and the MCP server at trust.paladinfi.com are proprietary; this repository covers the client-facing spec and examples under MIT.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。