What you receive
- Two API keys, one for production and one for testing, on the same account. Authentication is
Authorization: Bearer; no OAuth is needed. Keys rotate without downtime, because the old and new key work side by side, and a revocation takes effect from the next request. On request, keys can carry an expiry date. - The Partner plan:
| Requests per minute | from 300; more once we see your actual traffic |
| Concurrent requests | 16 |
| Credits per month | by agreement |
| Tools | every tool except bulk export |
- MCP or REST, with the same key. MCP Streamable HTTP for chats that speak MCP, and REST under
/v1/for everything else, documented at mcp.bloki.si/docs. - The quota on every response: in
_meta.quotaover MCP, in theX-Quota-*headers over REST. - Structured refusals: each gives a code and a reason, and the code says whether a retry makes sense. A refused request costs no credits.
- A monthly usage report: credits by tool, the number of your end users and the heaviest among them.
- An integration guide in Slovene: endpoint, headers, limits, error codes and a connection check.
Steps
-
Write to info@bloki.si. Tell us what your product is and who uses it, whether your chat speaks MCP or will call REST, how much traffic you expect (sessions a day and tool calls per session), and who your technical contact is.
-
We agree on the plan and the terms: limits, monthly credits and the price per credit. Billing is by credit, not by request: tools cost 1 to 25 credits, and each one’s cost is on the tools page and in the
account_statustool. -
You receive the keys over a secure channel, never in clear by email. A key is shown only once, when it is issued, so store it straight away. Keep it on your server only: never in a browser, a mobile app or a model’s prompt.
-
Connect. The MCP endpoint ends in a slash; without it the server answers 307, which some clients do not follow.
MCP: https://mcp.bloki.si/mcp/ REST: https://mcp.bloki.si/v1/… Authorization: Bearer bloki_sk_live_… -
Send
X-Bloki-End-Userwith every request: an opaque identifier of the end user (up to 64 characters ofA–Z a–z 0–9 _ -). The same user always gets the same value. Never send an email address, a name or anything else that identifies a person; we recommend an HMAC-SHA256 of your internal user id with a secret only you know. That way one user automating your chat does not stop all the others. -
Put the server’s instructions in your model’s system prompt. The server sends them when you connect. In short: resolve first (
find_area,find_address,resolve_property), identifiers are always strings, and a statistic comes with itsn. Two ready-made prompts,analyse_addressandarea_brief, are available too. -
Handle the quota and refusals. Too many requests in a minute: wait and retry. Monthly credits used up, or a tool outside the plan: do not retry. A content refusal, where the register does not answer the question, is not an error but an answer to pass on to the user.
-
Check the connection:
curl -si https://mcp.bloki.si/v1/server-info \ -H "Authorization: Bearer $BLOKI_KEY" \ -H "X-Bloki-End-User: test" | grep -i '^HTTP\|^x-quota'Expect
HTTP/2 200withX-Quota-*headers; a 401 means a wrong or revoked key.
Terms
Your model sits between the server and the user, so the agreement carries five rules:
- Caveats are passed on. Caveats, sample sizes (
n), thin-data flags (thin) and content refusals are not removed or hidden from the user. A summary is fine if it keeps their meaning. - Credit the source: “Source: bloki.si (GURS, ETN and other public registers)”.
- Personal data. Private owners cannot be identified, and the server does not return a sole trader’s address by name. These are properties of the server, not settings, so you do not promise your users otherwise.
- Not a valuer’s appraisal. Value estimates are statistical, with a measured error, and are not presented as an appraisal by a licensed valuer.
- Keys are confidential. If you suspect one has leaked, tell us at once and we revoke it.
Request an integration
Write to info@bloki.si with the subject “Integration”. We reply with the guide and a proposed plan.
For your own use you need no integration: connect your AI assistant for free in two minutes — see setup.