Blocklore's claims are open to programs as well as people. Everything here is read-only except
suggestions, needs no key, and allows any origin (CORS). The base is
https://blocklore.space/api/v1.
Blocklore runs no model for readers. To have an assistant answer from the claims, connect it to the MCP server; /connect shows the setup for each client.
Every answer carries an edition: the block height and time the store was compiled at, and the
build. Two answers with the same edition come from the same store. Each claim's url is the
address to cite.
The store is licensed CC BY 4.0 (knowledge/LICENSE): credit Blocklore and link each claim you use, for example "Blocklore, https://blocklore.space/claims/, CC BY 4.0". The short quotes inside sources belong to their authors and are included as citations. The code is MIT.
Endpoints
| Path | What it returns |
|---|---|
GET /api/v1 | This list, the edition and the license note. |
GET /api/v1/claims | Every published claim in brief: id, url, statement, kind, topics, depth, status, confidence, last verified, counts of sources and checks. |
GET /api/v1/claims/{id} | One claim's whole record: sources with locators and quotes, checks, every revision with what it said before, and where it is shown. |
GET /api/v1/topics | Every published topic: title, group, lede, claim count, how many claims wait for review. |
GET /api/v1/topics/{slug} | One topic: its sections in order, each claim in brief, and the concept to code ladder of its sources. |
GET /api/v1/changes?since={height}&limit={n} | Change events newest first. since keeps only events stamped after that block height; limit is 1 to 500, default 100. |
POST /api/v1/search | Claims ranked by BM25 for a query. See below. |
GET /api/v1/queue, POST /api/v1/queue | Reader suggestions: the questions the brain or the owner has triaged and how many wait; suggest a question for research. See below. |
GET /api/v1/bips | The BIP index at a commit, each BIP's link at that commit, the claims citing it, and every change seen since the first reading. |
GET /api/v1/snapshots | Every snapshot, newest first: block, commit, root, previous root, count, and what its OpenTimestamps proof says (the blocks it reaches or the calendars still holding it), with links to both files. |
GET /snapshots/{n}.json, GET /snapshots/{n}.json.ots | A snapshot file and its proof, byte for byte as committed. See /verify. |
GET /api/v1/gaps | The gap map: per topic, the questions from Blocklore's test set of 200 with known answers that the published claims cover only in part or not at all (placed on the nearest topic by search), depths with fewer than 3 claims, failed rechecks and claims waiting for review, most open gaps first; plus questions near no topic and search misses. See /gaps. |
GET /exports/claims.jsonl | Every published claim's whole record, one JSON object per line, in id order. |
GET /llms.txt | A short index for language models (llmstxt.org). |
GET /llms-full.txt | Every published claim as Markdown, topic by topic, with its sources and quotes. |
An unknown id or slug answers 404 with {"error": "..."}. Claims held for human review are not in
any answer until the owner approves them; topics report only how many wait.
Search
Search takes its text in a POST body, so the text never lands in a URL that servers and proxies write to their logs.
curl -s https://blocklore.space/api/v1/search \
-H 'Content-Type: application/json' \
--data '{"q": "taproot key path", "limit": 5}'
The body is {"q": text, "limit": 1 to 50}, at most 2000 characters. GET answers 405 with this
explanation. Results are current claims in brief (superseded ones are left out), each with a
score. Ranking is BM25 with prefix matching and light fuzzy matching for longer words; opcodes
such as OP_RETURN are one term. The same query on the same edition returns the same list.
A query that looks like a seed phrase, a private or extended key, a descriptor with keys, an address, a 64-character hex string (a key or a transaction id), a Lightning invoice, a Nostr secret, an email address or a phone number is refused with 400, and it is neither echoed nor kept. Blocklore never needs your keys; if you pasted real ones, treat them as exposed.
Suggestions
Anyone can suggest a question for Blocklore to research, here or in the box on /gaps:
curl -s https://blocklore.space/api/v1/queue \
-H 'Content-Type: application/json' \
--data '{"q": "How do statechains move bitcoin between owners?"}'
The body is {"q": text of at most 300 characters}, at most 2000 characters in all; any other
field is refused with 400. No model reads it on the site. A question that looks like it holds a
seed, key or address is refused with 400, like search. Links, bare domains, email addresses, Nostr
keys and handles are removed, and what is left must be at least 8 characters. Suggesting a
question again counts rather than adds. Each visitor can suggest 5 a day and the whole site takes
200 a day; past either, the answer is 429 until 00:00 UTC. The result is
{"queued": true | false, "item"}: queued is false when the question was already there and
now counts once more.
GET /api/v1/queue lists the questions the brain or the owner has triaged (researching or
answered, with the claims that answer it, and a note) and waiting, how many are not triaged
yet. Untriaged and declined suggestions are never shown. An item's missing field is null for
suggestions; it held Ask's note on questions queued before Ask was removed.
MCP
https://blocklore.space/mcp is a Model Context Protocol server, so an assistant can search and
cite the store as tools. It reads the same store as this API and names the same edition. It is
read-only and needs no key or account.
claude mcp add --transport http blocklore https://blocklore.space/mcp
Claude (desktop and web) takes the same URL as a custom connector, Cursor and VS Code take it in
their mcp.json, and any other client takes it as a remote (HTTP) MCP server.
/connect shows each setup.
| Tool | What it does |
|---|---|
search_claims | query (up to 200 characters), limit 1 to 50: claims in brief, best first, as POST /api/v1/search ranks them. |
get_claim | id: one claim's whole record, as GET /api/v1/claims/{id}. |
list_topics | Every published topic, as GET /api/v1/topics. |
get_topic | slug: one topic's sections and claims, as GET /api/v1/topics/{slug}. |
list_changes | since_height, limit 1 to 500: change events newest first. |
get_bip | number: one BIP from the index, the claims citing it, and its changes. |
list_bips | status, layer (both optional): the BIP index, filtered. |
Every result carries structuredContent and the same JSON as text. A lookup that finds nothing,
or arguments that do not fit the schema, come back as a tool error the model can read. Every
string argument goes through the secret check above before any tool reads it.
The server speaks protocol revision 2026-07-28, in which each request carries its version and
capabilities in _meta and mirrors them in the MCP-Protocol-Version, Mcp-Method and Mcp-Name
headers. It also serves clients that open with initialize (2025-11-25, 2025-06-18, 2025-03-26),
without minting a session. Every answer is one JSON object; no stream is opened, and GET
answers 405.
Examples
What changed since block 976,700:
curl -s 'https://blocklore.space/api/v1/changes?since=976700&limit=20'
One claim with its evidence:
curl -s https://blocklore.space/api/v1/claims/taproot.activation-height
Everything at once, for your own index:
curl -s https://blocklore.space/exports/claims.jsonl > blocklore-claims.jsonl
Stability and use
Version 1 changes only by adding fields and endpoints. A field is never renamed or given a new
meaning inside v1. One exception, by the owner's decision on 2026-11-21: Blocklore stopped running
a model for readers, so POST /api/v1/ask and GET /api/v1/scorecard are gone (404), and
POST /api/v1/queue takes {"q"} alone (the Ask ticket it used to require is refused with 400).
Answers are cached for up to five minutes and change when a new edition deploys.
Reading has no rate limit today; please fetch the export rather than every claim one by one. A paid tier for heavy automated use may come with Lightning payments (L402); reading as a person will stay free.