Connect an agent
Regulatory Terminal: payment and e-money licensing across jurisdictions, with the register entry or the quoted provision behind every value. Which regimes exist, what each permits and for which client types, how far it passports, who holds it, and where a provider's claimed coverage cannot be traced to an authorisation. Read-only. Every value carries its tier, quote, source URL and retrieval date, and absence is reported as absence rather than as zero.
The endpoint
Public and free while in beta. It speaks MCP over streamable HTTP, needs no account, no key and no sign-in: paste the URL into your client and the tools appear. The corpus is read-only, so no tool here can change anything.
The only limit is 120 tool calls per minute from one address, which no person typing questions will reach; past it the server answers 429 with a Retry-After header. It runs around the clock on the same data as these pages, refreshed from the registers as they change. If it is down, /api/health says so.
What the agent gets
6 read-only tools. Every value they return carries the tier it was asserted at, the verbatim quote, the source URL and the date it was read, or an explicit statement that the corpus does not carry it. An agent using this can tell its own user where a number came from, which is the whole reason to prefer it to a search result.
- search
- Search the regulatory graph. Find the id of a jurisdiction, regime, company, product or statute by name.
- get_jurisdiction
- Jurisdiction: regimes, inbound passporting, local holders. Everything the corpus holds about one jurisdiction: the regulators, every local licensing regime with its minimum capital and authorisation timeline, which regimes elsewhere passport into it, and which companies hold what locally.
- get_regime
- Regime: requirements, grants, passporting reach, holders. One licensing regime in full: the statute behind it, every requirement field the corpus has resolved, which activities it grants and to which client types, which jurisdictions it passports into, and who holds it.
- check_coverage
- Coverage for a company, claimed versus verified. Whether a named company can actually serve an activity in a set of jurisdictions, and by what route: its own licence, a passport, or a rented one through a provider.
- plan_footprint
- Cheapest route to a target footprint. The inverse query: name the markets, products and client types you want to serve, and this returns the combinations of licences, passports and rented routes that reach them, on the cost frontier.
- get_evidence
- The full record behind one claim id. The full record behind one claim id: the verbatim quote, the source name, kind and URL, the tier, the retrieval date, and whether the document is still reachable.
By client
https://terminal.szyma.co/api/mcp
claude mcp add --transport http regulatory-terminal https://terminal.szyma.co/api/mcp
{
"mcpServers": {
"regulatory-terminal": {
"url": "https://terminal.szyma.co/api/mcp"
}
}
}{
"servers": {
"regulatory-terminal": {
"type": "http",
"url": "https://terminal.szyma.co/api/mcp"
}
}
}{
"mcpServers": {
"regulatory-terminal": {
"serverUrl": "https://terminal.szyma.co/api/mcp"
}
}
}https://terminal.szyma.co/api/mcp
If you are building against it
No authorization step: the endpoint answers any MCP request directly, and /.well-known/oauth-protected-resource returns 404, which clients read as "no auth required". Send initialize, then tools/list and tools/call as usual.
Protocol revision 2026-07-28. Requests carry the per-request _meta envelope and the Mcp-Method and Mcp-Name headers; a client library current enough to advertise that revision does this for you.