Current MCP tool reference

The current Research MCP exposes 28 tools: 14 data, analysis and help tools, and 14 focused display tools. It uses the same Research API as the browser product. The retired publication-oriented tool catalogue is not the current API.

Connect to https://mcp.thecoindaily.co/mcp using OAuth. Research and MCP access require a current subscription or an applicable complimentary pass.

Data, analysis and help

Tool	Use it for	Inputs and follow-up
research_capabilities	Discover supported workflows and limits without fetching market data	Browse area, optionally filter short keywords; use returned case IDs and pagination
research_discover	Resolve canonical tickers and governed group membership	Asset query, exact returned group_id, offset, limit; inspect groups to resolve category labels
research_markets	Read the prepared market board in published Signal Book order	Optional ticker substring, offset and limit; rank is not a predicted return
research_open_ticker	Retrieve price candles and the available layer/study catalogue	Exact ticker, UTC window, candle timeframe 1h, 4h or 1d
research_read_series	Retrieve one selected metric or study	A request with ticker, series, dates and any required returned variant/indicator/venue IDs; follow its cursor
research_market_context	Read hourly model regime and observed volatility for one ticker window	Exact ticker and an explicit UTC window; readiness, gaps, source pair and historical percentiles stay explicit. See Market regime and volatility
research_interpret_range	Calculate prominent-move findings and selected supporting metrics	A request preserving the original A/B candle times, timeframe and enabled layers; deterministic Research calculation, not an LLM call
research_prepare_cohorts	Prepare governed comparison history	days: 1, 7 or 30; waits briefly for a cold build and returns the snapshot (no job id) with its initial comparison and facts.summary
research_cohort_status	Legacy compatibility: check a pre-existing preparation job	Exact job id only when already obtained from a pre-deploy response or a direct Research API call; prepare does not return one
research_compare_cohorts	Compare chosen groups on a pinned history	A request with returned snapshot/group IDs, bounds and comparison controls; preserve unchanged controls on refinements. Returns facts.summary headline returns, relative percentage points and charted/member counts
research_briefing	Read the latest published Market Update and chapters	No selection arguments; publication readings are not live prices
research_ticker_videos	Find ticker-primary videos by publication time	Exact ticker and UTC window
research_news_context	Retrieve linked stories and daily attention around a day	Exact ticker and explicit anchor day; use the actual window returned by the API
research_read_evidence	Inspect exact sections of an already captured result	evidence_id, JSON-pointer path, offset and limit; does not refresh the data

Use the server's tools/list schemas for exact request fields, enums and bounds. Do not manufacture group IDs, study variants, job IDs or evidence IDs. research_capabilities explains workflows in machine-readable form.

Display tools

All display tools require the exact evidence_id from a compatible preceding result. Evidence belongs to the authenticated principal and is normally retained for up to 15 minutes. The released shared store is restart-safe: a routine service replacement does not by itself discard an unexpired capture. Retained old widget resources, current tool metadata and expiring market evidence are separate states; a published widget resource can remain usable while a new result still needs a fresh read.

Display tool	Compatible retrieval
research_show_markets	research_markets
research_show_directory	research_discover
research_show_price	research_open_ticker
research_show_funding	research_read_series — funding
research_show_open_interest	research_read_series — open interest
research_show_liquidations	research_read_series — liquidations
research_show_liquidity	research_read_series — liquidity
research_show_positioning	research_read_series — positioning
research_show_market_context	research_market_context
research_show_study	research_read_series — a selected indicator
research_show_cohorts	research_prepare_cohorts / research_compare_cohorts
research_show_briefing	research_briefing
research_show_videos	research_ticker_videos
research_show_news	research_news_context

Cohort display supports performance, relative, contributions, relationships and correlation views. Copy returned metric IDs when selecting relationship axes. Briefing display can select an exact returned chapter ID.

Selection and time

• Chart and video windows use an inclusive UTC start and exclusive end, bounded to 31 days.
• Range interpretation selects A and B by their original candle opening times and includes the whole B candle. Do not pass an exclusive chart end as B.
• Indicators default to calculation=candles on the selected 1h, 4h or 1d interval. Use its exact catalogue code/variant and preserve warm-up/readiness/coverage. An explicit calculation=native keeps the stored source clock; there is no native fallback.
• Numeric funding/OI/liquidation/L2/account/whale reads default to alignment=candles at that same interval. Research selects snapshots or sums flow periods; retain observed_at, available_at, usable, comparison_eligible and bucket coverage. Partial flows are observed subtotals, not complete totals. Funding retains its percent-per-eight-hours basis; OI venue lines use the API's matching-cycle values without client sums. Explicit alignment=native retains raw source rows. Market context always uses its own hourly observation clock and has no candle-alignment mode.
• Range selections use selection.alignment=candles for the same snapshot/flow policy and include each study's id, indicator, calculation and timeframe.
• Cohort comparison uses a pinned snapshot. Its selected time bounds are Unix seconds, not JavaScript milliseconds.
• A video's publication time is not necessarily the market period it discusses. News context supplies its own actual lookback window.

Response and evidence

The structured result uses schema_version: tcd.mcp-research/v2. It identifies the surface, status, selection, compact facts, evidence reference and Research link. Full widget payloads are carried separately from the compact model-facing summary. Use research_read_evidence for exact retained sections before citing numbers absent from that summary; retrieve the specific section a question needs rather than enumerating sections.

Cohort results add facts.summary, derived only from the returned payload: the selected window start/end and read time; each group's return_pct, benchmark-relative relative_pp and as_of taken from the last returned line point (not necessarily an exact window end), plus its returned charted_count/member_count; and a benchmark object with the returned ticker, label, return_pct and relative_pp. A value the payload does not provide is omitted, not defaulted. When research_prepare_cohorts returns the comparison the user asked for, this summary answers the headline question without another comparison or evidence read. A comparison re-issues only the API's busy 429 signal, once; the six-second bound covers retry waiting, not total request time.

Status	Meaning
available	The requested result is available
partial	Some observations or coverage are incomplete; inspect the supplied detail
empty	No matching observations in the returned scope
preparing	An upstream preparation is still running; the requesting tool waits or returns retry guidance. A legacy cohort job id, if already known, can be checked with research_cohort_status
unavailable	A required source or service could not provide the result
expired	The retained reference is no longer usable; retrieve fresh evidence
error	The request or result could not satisfy the contract

Missing readings are not zero. Preserve source units, contributor coverage, timestamps and the selected window. A newly fetched range can contain source corrections; do not silently replace an earlier captured table.

Errors and access

An unauthenticated request receives 401. An authenticated identity without current access completes the MCP handshake and can list tools, resources and prompts. A data method is refused at HTTP 200 with a neutral access explanation (access information only — no checkout or upgrade link) that the assistant can show in the conversation: tools/call returns an MCP tool error result (isError: true) so hosts surface the text to the model, while resources/read, prompts/get and completion/complete return a JSON-RPC error. The informational help tool research_capabilities (static, versioned guidance with no market-data read) stays available without access so an assistant can explain what the product offers. No research data is served without access. An unavailable subscription check receives 503 and should be retried—not treated as a request to buy another plan. Rate limits may return 429.

See pagination, authentication and troubleshooting. This is a research interface with no wallets, trade execution or order placement.