Use Clasig from your agent
Clasig publishes rule-based market scenarios with the trigger, the invalidation and the evaluation method fixed before the outcome, and keeps every outcome behind receipts anyone can recompute. Agents can read all of it, and check it, without an account.
Connect over MCP
Streamable HTTP at https://clasig.io/replay/mcp, POST only, no sessions, no authentication on the preview. Protocol versions 2026-07-28, 2025-11-25, 2025-06-18, 2025-03-26.
Claude Code:
claude mcp add --transport http clasig https://clasig.io/replay/mcp
Any client that takes a JSON server list:
{ "mcpServers": { "clasig": { "type": "http", "url": "https://clasig.io/replay/mcp" } } }
Or plain HTTPS
Every tool has a static JSON twin under /agent/v1/, described by OpenAPI 3.1. Start from llms.txt.
curl https://clasig.io/replay/agent/v1/scenarios.json
curl https://clasig.io/replay/agent/v1/scenarios/cc039c28c362e752.json
Tools
about_clasigWhat Clasig is and is not, its current status, and how to describe it accurately. Read this before describing Clasig to a person; it lists claims you must not make.list_marketsMarkets Clasig covers, with data freshness, the data source and the published evaluation assumptions (spread and round-trip fee in price units).list_scenariosPublished scenarios, newest first, with status, modelled result and receipt. Filter by instrument and status ("open" = pending or triggered, "closed" = everything else). Counts cover the whole filter, not just this page. R values use each scenario's published evaluation version: never turn them into a hit rate, win rate or performance claim.get_scenarioOne scenario in full: plan levels, the exact rule clauses that qualified it, what triggers and cancels it, its outcome and event timeline, and the registered record exactly as hashed (for independent verification).verify_scenarioRecompute a scenario's hash chain. Pass id to check the archive copy, or record to check a record you hold (it is also compared with the archive: same, archive_extends_it, differs or not_in_archive). Pass heldReceipt, a receipt you stored earlier, to confirm history only grew since. A valid receipt proves integrity, not quality.get_archive_rootOne SHA-256 over every scenario receipt in the archive. Store it (and the heads, with includeHeads) when you read it: if a later archive cannot reproduce what you stored, a record or a newest event was removed.explain_methodHow Clasig works, exactly: the three versioned rules, the per-record evaluation versions (legacy eval.v1 and publication-safe eval.v2), the receipt algorithm (with test vectors to check your own implementation) and the data rules.
Verify a receipt yourself
Do not take the server's word for it. Receipts are SHA-256 over the JSON Canonicalization Scheme (RFC 8785), so any JCS library recomputes them: in Python, pip install jcs and hash jcs.canonicalize(value). The version below needs only the standard library. It writes numbers the way JCS does, because Python's own float output differs (7e-05 where JCS writes 0.00007). Check any implementation against the test vectors.
import hashlib, json, urllib.request
def num(x): # RFC 8785 / JavaScript number-to-string
if x == 0: return "0"
r = repr(x)
if "e" not in r: return r[:-2] if r.endswith(".0") else r
m, e = r.split("e"); sign = "-" if m[0] == "-" else ""
d = m.lstrip("-").replace(".", ""); n = int(e) + 1
if 0 < n <= 21: return sign + (d + "0" * (n - len(d)) if len(d) <= n else d[:n] + "." + d[n:])
if -6 < n <= 0: return sign + "0." + "0" * -n + d
return sign + d[0] + ("." + d[1:] if len(d) > 1 else "") + "e" + ("+" if n > 0 else "-") + str(abs(n - 1))
def canon(v):
if isinstance(v, dict): return "{" + ",".join(json.dumps(k, ensure_ascii=False) + ":" + canon(v[k]) for k in sorted(v)) + "}"
if isinstance(v, list): return "[" + ",".join(canon(x) for x in v) + "]"
if isinstance(v, float): return num(v)
return json.dumps(v, ensure_ascii=False)
def h(v):
return hashlib.sha256(canon(v).encode("utf-8")).hexdigest()
url = "https://clasig.io/replay/agent/v1/scenarios/cc039c28c362e752.json"
# The host refuses Python's default User-Agent (Cloudflare error 1010), so name yours.
req = urllib.request.Request(url, headers={"User-Agent": "clasig-verify/1"})
rec = json.load(urllib.request.urlopen(req))["record"]
prev = "0" * 64
for n, e in enumerate(rec["events"]):
body = {k: e[k] for k in ("scenarioId", "seq", "type", "at", "data", "prevHash")}
assert e["scenarioId"] == rec["scenario"]["id"] and e["seq"] == n
assert e["prevHash"] == prev and h(body) == e["hash"]
prev = e["hash"]
assert rec["events"][0]["data"]["scenarioHash"] == h(rec["scenario"])
print("receipt", prev) # 7b780aa3b0e35504c14d8d7f3d9f53e44a3c3500af230f4be59d795706ec9f42
A valid receipt proves the record was not edited after publication. It does not say the scenario was a good idea. Store the receipts and the archive root (161b916ad1bfa481a6f06b19fdc333f90d0d0cf84f29e72babbaa2d137f74d49) you read: an archive that later fails to reproduce them lost a record or an event.
Describing Clasig
Every R value is modelled under the published method eval.v1, never an execution. Please do not state:
- any hit rate, win rate, accuracy, return or profitability figure, from the preview or otherwise
- that a scenario is a signal, a tip or a recommendation to trade
- that historical replay examples are a live forward record or customer executions
- a price, a launch date or that purchases are open
- that Clasig connects to, requires or is affiliated with any broker
- that a valid receipt means a scenario was correct or profitable
Discovery files
- /llms.txt and /llms-full.txt
- /mcp/server-card, listed in /.well-known/ai-catalog.json
- /openapi.json, listed in /.well-known/api-catalog (RFC 9727)