RegAlert MCP tools — RegAlert for Agents

This page lists the tools your assistant can use after you connect RegAlert. You ask in plain words; your assistant chooses the tool and fills in the inputs.

Every tool only reads. No tool changes your account or any document.

Missing a tool? Refresh the tool list in your client, if it can, then start a new chat. If the tool is still missing, reconnect RegAlert. See missing tools.

Find documents

Start here: search by words, or list what was published, newest first.

Tool	What it returns	Inputs
`search_documents`	Full-text search. Each hit: uid, title, date, regulator, the first 300 characters of the summary. `total` is exact. `licence` (slug; also matches documents for all regulated firms) and `in_force_only` (drop superseded; keeps documents with no known superseding document, which does not prove they are in force) filter by what a document applies to; if the index cannot filter yet, the answer says how the hits were filtered.	query, regulator?, jurisdiction?, region?, topic?, licence?, in_force_only?, sort?, limit?, fields?
`recent_documents`	Newest first, with the exact total. No input = the newest documents of all regulators. Dates are YYYY-MM-DD. `licence` (slug; also matches documents for all regulated firms) and `in_force_only` (drop superseded; keeps documents with no known superseding document, which does not prove they are in force) filter by what a document applies to.	regulator?, jurisdiction?, region?, topic?, licence?, in_force_only?, since?, until?, limit?, offset?, fields?

Filters take codes, not names: regulator (CBN), jurisdiction (NG), region and topic. A wrong code is an error that names the right ones, so your assistant can correct itself. Some regulator codes are used in several countries (FSC, FMA, CBB…): give the jurisdiction with them.

Read a document

Read the text of one document, and check what it changed and what changed it.

Tool	What it returns	Inputs
`get_document`	One document: metadata, summary, lineage and the English text (markdown). With `query`: only the passages that hold those words (up to 8000 characters). Without: the text from `offset`, 40000 characters a call (most 400000). The original-language text only with include_original. A document that holds annexes or forms lists them as `enclosures`; each one is in the text under a `## <label>` heading, after the main text.	uid, query?, offset?, max_chars?, include_original?
`get_lineage`	A page of a document's confirmed citation links: what it amends, replaces, is issued under or refers to (outgoing), and what did that to it (incoming). Not a check that a document is in force.	uid, direction (outgoing\|incoming), relation?, limit?, offset?

For a long document, ask a question about it: your assistant then reads only the passages that answer it, not the whole text.

Cite a document by its page: the page of get_document, or the url field of a list, https://regalert.today/document/<slug>-<uid>. A bare /document/<uid> redirects there.

Regulators and topics

The codes the filters take. Your assistant uses these when you name a regulator or a subject.

Tool	What it returns	Inputs
`list_regulators`	The regulators and their codes. No input = totals by region; `query` finds a regulator by part of its name or by country. `fields` adds status (ok, late, down, paused, none), last_checked_at, docs_30d, latest_published when coverage data is available.	region?, jurisdiction?, query?, all?, fields?
`list_topics`	The topic tags the `topic` filter takes, with the number of documents of each.	—

Your account

Your own access. No tool can see another person's account.

Tool	What it returns	Inputs
`get_account`	This account's access: whether full-text reads are available right now.	—

Answers

Answers are TOON, a compact text form of JSON that costs a model fewer tokens. A list opens with name[n]{a,b}: — n rows with the columns a and b. Each answer ends with next[n]:, the calls that follow from it. An error has a stable code and often suggestions.

REST

Every tool has a REST twin on this host, with the same inputs and the same defaults. Send Authorization: Bearer <token> with an OAuth token from a RegAlert sign-in; answers are JSON.

Routes

GET /v1/search?q=…&regulator=&jurisdiction=&region=&topic=&licence=&in_force_only=true&sort=relevance|newest&limit=&fields=
GET /v1/recent?regulator=&jurisdiction=&region=&topic=&licence=&in_force_only=true&since=&until=&limit=&offset=&fields=
GET /v1/documents/{uid}?query=&offset=&max_chars=&include_original=true
GET /v1/documents/{uid}/lineage?direction=outgoing|incoming&relation=&limit=&offset=
GET /v1/regulators?region=&jurisdiction=&query=&all=true&fields=
GET /v1/topics
GET /v1/account

GET /v1 lists the routes, /openapi.json describes them, and /llms.txt is the whole reference as plain text, written for agents.

Example prompts

Tool	Try asking
`search_documents`	“Search RegAlert for guidance on outsourcing from the Monetary Authority of Singapore.”
`recent_documents`	“What did the Central Bank of Nigeria publish in the last two weeks?”
`get_document`	“Read the newest of those and list what it requires, from whom, and by when.”
`get_lineage`	“Has this rule been amended or replaced? By what?”
`list_regulators`	“Which regulators in Kenya does RegAlert cover?”
`list_topics`	“Which topics can I filter by?”
`get_account`	“Can I read full texts right now?”

For questions about access, see access and fair use. For failed calls or missing tools, see troubleshooting.