Investor Intel for agent builders: candles and whole series for your agent, and an Agents page that shows its working

Candles for your agent

get_candles returns one asset's open, high, low, close and volume at any of the chart's fifteen widths, from one minute to one calendar month, over any of its ranges. It reads the same stored candles the asset page draws and never calls a provider. Every answer says the width served beside the width you asked for, which provider supplied each candle, and every candle left out, whether by your plan's price history allowance, the chart's display ceiling or the page of 300, with a pointer to the next page. The public demo server offers it too, without CoinMarketCap candles, which are for signed-in members.

Whole series, in long form

export_series returns a whole stored series in pages, oldest first, as JSON or CSV: one asset's price history, every hourly market regime capture, or every new listing in a window. Nothing is sampled. Every row carries a figure_owner column, investor_intel for figures we calculate and the provider for theirs, so a pipeline can keep what it may republish apart from what it must ask about first.

The token verdict for your agent

token_verdict gives your agent the token page's verdict panel for one contract: exit liquidity, the top 10 and developer shares, bundlers and snipers, LP burn or lock, mint and freeze authority, tax and honeypot, each fact with its source and its time, from the same stored readings and the same rule the page uses. A contract nobody has opened says it has no stored reading rather than no risk. The priced exit check stays on the token page, because it is a live quote. For members.

The same reads through the agent API

The agent API answers read_candles and export_series with the same tools, so both ways in give the same answer.

A setup checklist on the Agents page

The three steps from onboarding now stay on the Agents page, ticked from your own tokens: create a token, test it, make a first call, with the first call and a sample newsletter prompt beside them.

Only the permissions the hosted server honours

The token form offers the permissions a hosted tool uses and says which tools each one opens. Portfolio is no longer ticked for you. Portfolio, Theses and Charts, which only the local bridge and the agent API use, sit one click away under their own heading, and the local bridge has its own page in the docs.

Revoking asks first

Revoke now names the token and asks you to confirm, with Keep it beside the button, because a revoked token cannot be brought back.

Test a token in full

Test this token shows the request it sent, with the token cut to its first characters, the HTTP status, the headers and the time taken, and how many tools came back. Test and run whoami also shows the token's permissions and expiry, your plan and the calls left today, for one call. The token is cleared from the field as soon as the request goes.

The demo's shared limit, stated

The Agents page now says that every visitor to the public demo server shares 10,000 tool calls a day, so a refusal near the end of a busy day may not be yours.

Easier for an agent to find

llms.txt now names the MCP servers and the Agents page, and /.well-known/mcp.json describes both servers, how each one authenticates and where its documentation is. The site footer links the Agents page.

A better sample newsletter prompt

The onboarding prompt now reads the week in candles, keeps only inspected listings with real volume, and says how many listings were not inspected.

Developer documentation, rewritten

The agent access and hosted MCP pages in the docs are now written for the people using them, without internal notes, and no longer state tool counts that drift.

Tools that arrived without a note here

Between September 20 and October 24 these tools joined the server and were not described in this changelog until now.

• market_structure: the precomputed market figures the Structure pages show, one figure a call
• rwa_best_wrapper: which wrapper of a tokenized asset is cheapest to its anchor, which is closest, and which trades most
• rwa_coverage: how much of the tokenized asset universe has a token that actually trades
• rwa_universe_changes: what entered and left the universe, day by day
• rwa_issuer_concentration: how concentrated the issuers of each asset type are
• rwa_premium_history: how a wrapper's premium to its anchor moved over time
• rwa_exit_capacity: how many days a position takes to sell
• alert_create_rule: proposes an alert on any other trigger the Alerts page offers, which you approve
• daily_brief_read, explain_read and saved_research_read: read back the briefs, Explain This answers and saved research Investor Intel already wrote for you

One tool count everywhere

Pages, docs and this changelog stated the number of tools four different ways. Every count is now read from one list: 44 tools, 19 of them in the public demo.

Two earlier notes corrected

The September 20 entry said a trial had no daily ceiling on agent calls; the trial allows what Pro allows, 10,000 a day. It also said /intel/markets/bitcoin opens Bitcoin; it does when the address names one asset exactly, and when the address also matches a token that carries BITCOIN as its ticker you choose, with the largest first.