Pick your client, then add it in one click or paste the config.
No confirmed install command yet
We only publish a config when the install details come from the project itself — its README, its docs, or a verified owner. We haven’t found those for SanctionsKit developer MCP, and we’d rather show nothing than a guess you’d paste into your client.
Build SanctionsKit integrations from your editor. Find API documentation, check request formats, and test invented subjects against the sandbox.
The plugin bundles the hosted SanctionsKit MCP server connection and an integration skill for Codex, Claude Code, and Cursor. There is no local server to run and no install script.
A separate Gemini CLI extension connects to the four public documentation, schema, and request-validation tools. It requires no SanctionsKit account or API key and does not expose sandbox account tools.
Clone this repository and open its directory. Install with the instructions for your client below, then try:
Find the SanctionsKit screening request schema and validate an invented example without submitting it.
Documentation, schemas, and request validation work without a SanctionsKit account. Sandbox account tools require a connection to a SanctionsKit workspace. Each client's OAuth registration must be configured on the hosted service; see authentication setup.
Connect to the hosted MCP server
You can also connect directly from an MCP client using Streamable HTTP at:
text
https://www.sanctionskit.com/mcp
Public documentation, schema, and validation tools need no credentials. For sandbox account tools, sign in through the client's OAuth connection flow after its client ID and callback have been registered with SanctionsKit. The requested scopes are sources:read, screenings:write, results:read, and usage:read. See authentication setup for client configuration and verification status.
Codex
From the repository root:
sh
codex plugin marketplace add .
Open the plugin directory in Codex, choose the SanctionsKit marketplace, and install SanctionsKit. Start a new task after installation. For a GitHub installation, pass this repository's HTTPS GitHub URL to the same command instead of ..
You can also give /plugin marketplace add this repository's HTTPS GitHub URL. Restart Claude Code after installation. Use /mcp to inspect the connection and sign in when account tools are needed.
Cursor
For a local installation, enable Allow Local Plugin Imports in Cursor, copy the plugin folder into its local plugin directory, then reload Cursor. Enterprise policies may require an administrator to allow local imports.
This command is for a first install; replace the existing sanctionskit folder when updating. Use a real copy, since Cursor skips external symlink targets. Team administrators can use Import from Repo in their team's plugin marketplace with this repository's GitHub URL. A public Cursor directory listing requires a separate submission.
Restart Gemini CLI, then use /mcp to inspect the sanctionskit connection. Try:
Find the SanctionsKit screening request schema and validate an invented example without submitting it.
The root gemini-extension.json uses Streamable HTTP and allows only search_docs, get_doc, get_api_schema, and validate_screening_request. It includes no credentials, OAuth configuration, local executable, or bundled context file. Request validation checks structure only; it does not submit a screening or check a real subject. Gemini CLI's own access requirements still apply.
Sandbox screening and other account tools are not part of this extension. The integration skill bundled for the other clients is not installed by the Gemini extension. If you already configured a server named sanctionskit in Gemini's settings.json, that configuration takes precedence; remove or rename it if you intend to use this extension's public-tool configuration.
What you can do
Tool
Purpose
Connection
search_docs
Find integration guides and document slugs
Public
get_doc
Read a document from the published catalog
Public
get_api_schema
Inspect REST operations, schemas, and examples
Public
validate_screening_request
Check request structure without submitting it
Public
list_sources
Inspect source metadata and availability
Sandbox workspace
run_sandbox_screening
Submit invented subjects to sandbox@1
Sandbox workspace
get_screening_result
Retrieve retained sandbox evidence
Sandbox workspace
get_usage
Check sandbox allowance and reservations
Sandbox workspace
get_profile
Identify the connected profile
Sandbox workspace
The bundled sanctionskit-integration skill covers request design, validation, error handling, retries, and interpreting sandbox results. It uses the current tool schemas and documentation instead of assuming an SDK or endpoint exists.
Try these next:
“Show me how to add SanctionsKit screening to my server-side TypeScript app.”
“Validate a sandbox@1 request for an invented organization.”
“Run the synthetic Alex Morgan example, then show the result and remaining sandbox allowance.”
“Explain how I should retry a request after a timeout.”
MCP account tools operate in the sandbox. Production keys are rejected. A sandbox screening consumes sandbox allowance, retains evidence according to its retention policy, and may create a review case. Use a new idempotency key for a new intended screening; keep both the key and payload unchanged when retrying a lost response.
Validation checks structure only. Source catalog entries do not change the sandbox's synthetic coverage. A synthetic result says nothing about a real person or organization, and a no-match result is not legal clearance. Production integrations use the server-side REST API and your organization's review process.
Troubleshooting
Problem
What to check
Plugin does not appear
Add the repository root as the marketplace, install the plugin, and start a new session.
Reconnect the sandbox workspace with the required scopes.
Port already in use
Close the other login flow using that port, then retry.
Production key rejected
Use a sandbox connection for MCP; keep production integration in the REST API.
Result has expired
Report the expiry; create a new screening only if a new run is intended.
Coverage or allowance error
Inspect the returned error and current sandbox usage. Do not switch coverage silently.
Development
Node.js 22 or later is needed only for the repository checks. There are no npm dependencies.
Terminal
npm ci
npm run check
npm run smoke
check validates package structure and runs offline tests. smoke checks the public hosted MCP endpoint and its authentication challenge without signing in or creating screenings. A passing smoke check does not establish that a client's interactive sign-in works.
text
.agents/plugins/ Codex marketplace
.claude-plugin/ Claude Code marketplace
.cursor-plugin/ Cursor marketplace
gemini-extension.json Gemini CLI public-tool extension
plugins/sanctionskit/ Installable plugin, client configs, skill, and assets
docs/ Authentication and release guidance
scripts/ Package validation and public connection check
tests/ Offline verification
What it can reach and how reliably it runs — from our own checks, not the maintainer’s claims.
ActiveLast checked 2d ago
Recent health check succeeded.
Reviews
No reviews yet — be the first to share how it worked for you.
Frequently Asked Questions about SanctionsKit developer MCP
We don't have a confirmed install command for SanctionsKit developer MCP yet, so we don't publish a generated one — a guessed package name would point at the wrong package or none at all. Follow the project's own README or setup instructions (https://github.com/SanctionsKit/sanctions-kit-mcp-plugin) for the current steps.
Sanctions API docs and request validation, with account-linked synthetic sandbox testing.
We don't have confirmed pricing for SanctionsKit developer MCP. Listing it on AllMCPs is free, but that says nothing about what the server itself costs — check the project's own site for current pricing.
Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.