Skip to content

Kote Charts MCP

An assistant asked where Bitcoin sits in its cycle will answer from a number it cannot defend. The metric is either wrong, or right and unsourced, and in both cases the operator has no way to check it. The gap is not the model. It is that on-chain metrics travel without their definitions: an MVRV Z-Score means one thing at one vendor and something slightly different at the next, and the units, the update cadence and the known blind spots are exactly what a caller needs before quoting the number in a report.

Kote Charts is the version of that answer that ships its own working. 114 charts in 13 categories, computed from the vendor's own Bitcoin full node, free and keyless, and every one of them carries the formula, the payload fields with units, the input categories, the update cadence, the known limitations, the analysed block height and a ready-made citation string. The pass below drove all four tools through 17 tools/call requests, called both prompts, listed the resource surface, and reproduced one chart's figures from the other's.

Server type:   remote Streamable HTTP
Endpoint:      https://kotecharts.com/api/v1/mcp
Server:        kote-analytics, version 1.1.0, title "Kote Charts - Bitcoin on-chain analytics"
Protocol:      2025-06-18
Auth:          none (60 requests a minute per IP); an optional free key raises the REST tier to 3 requests a second
Tools:         4 (list_charts, describe_chart, get_chart, get_latest)
Resources:     116 (catalog, methodology, and one per chart)
Prompts:       2 (explain_metric, cycle_position_brief)
Registry:      com.kotecharts/charts, versions 1.1.0 and 1.1.1 published
Repo:          github.com/kote-charts/kote-charts (code MIT; data under separate terms)
Listed on:     the mcp.so submission queue as #7473

The evidence tier

Keyless and driven. No credential was needed at any point. initialize answered 200 as kote-analytics 1.1.0 on protocol 2025-06-18, tools/list returned four closed schemas, resources/list returned 116 entries of which 114 are per-chart methodology documents, and prompts/list returned two. Calls made: list_charts twice (once bare, once with query: "sopr"), describe_chart on mvrv-z-score, get_latest on mvrv-z-score and on etf-flows, get_chart on a full history, on a date range and on weekly granularity, plus four deliberate failures run to see whether the server refuses or invents.

The tool surface

Tool Arguments What it returns
list_charts query (optional substring, 80 characters) Every public chart with id, title, one-line blurb, category, formula, plus the total count. query: "sopr" narrowed 114 charts to 4 (adjusted-sopr, sopr, sopr-ratio, sth-lth-sopr).
describe_chart id (required) The metric's formula, what it shows, how to read it, caveats, fields with units, input data categories, update cadence, known limitations, data window, analysed block height, methodology URL, metric URL and chart page URL.
get_chart id (required), from, to, granularity, includePartial, exchange, maxPoints Historical rows shaped {t, ...fields} with the units defined by describe_chart.
get_latest id (required), exchange The most recent point plus the previous point, wrapped in an asOf block carrying lastT, publishedAt, processedAt and analysedBlockHeight, and a cite string.

114 charts, and the category count closes

list_charts returned 114 charts, and the per-category counts sum to that total exactly:

Category Charts Category Charts
Hot Charts 25 Realized 15
Flows & Whales 23 Valuation 11
Holders 7 Cross-Asset 7
Coin Age 6 Network 5
Supply & Issuance 4 Value & Volume 4
Addresses 3 Fees 3
Entities 1 Total 114

The submission advertises 110+ charts, so the count is honest and slightly understated. The set covers the metrics an operator would ask for by name (MVRV and its deviation bands, NUPL, SOPR and its adjusted and long-term variants, realized cap and realized price by age band, coin days destroyed, dormancy, Puell multiple, Mayer multiple, NVT, S2F reversion, SSR, spot ETF flows and ETF market cap against new supply) and the ones that exist because an analyst needed a comparison rather than a level (dxy-vs-btc, btc-gold, tether-supply-yoy, spot-vs-derivative-volume).

describe_chart is the reason a number can be quoted

describe_chart on mvrv-z-score returned the working, not just the series:

  • Formula and family: market cap minus realized cap, divided by the standard deviation of market cap; data family realized, source described as mix ยท onchain + price, onChain: true.
  • Fields with units: mvrvZ and mvrvZMa365 in zscore, usd in usd. That is the difference between a table that can be labelled and a table that has to be guessed at.
  • Cadence and coverage: updateCadence is daily (closed UTC days), granularities are day, 3day, week, month and year, and the data window runs from 2009-01-03 to 2026-11-16 on block height 976,137.
  • Limitations, in the vendor's own words: cost basis uses the daily close of the day each output was created rather than the exact trade price, lost or burned coins still count at their creation price, exchange-internal and off-chain transfers are invisible, and BTC/USD prices for 2009 and 2010 are sparse and partly reconstructed so early price-based values are approximate.

A limitation list that names its own weakest interval is worth more than a longer feature list, because the operator can now write the caveat into the note instead of discovering it in a review.

Two tools, one number

The pair that has to agree is get_latest and get_chart, and it does. get_chart with from: 2026-11-01 and to: 2026-11-16 returned 15 rows closing on 2026-11-15, and the final row is byte-identical to the latest block of get_latest:

Field get_latest Last row of the range
t 2026-11-15 2026-11-15
usd 64,119.34 64,119.34
mvrvZ 0.335 0.335
mvrvZMa365 0.723 0.723

The previous point was 2026-11-14 at 0.466 with the price at 68,232.31, so over one day the price fell 6.03% (68,232.31 down to 64,119.34) while the z-score fell 0.131 and the one-year average barely moved (0.726 to 0.723). An operator quoting Bitcoin is not in a euphoric regime now has the number, the definition of the number, and the date it is true for.

A second chart, checked by subtraction

get_latest on etf-flows returned the spot ETF line for 2026-11-13: netFlowBtc of -587.1 against a cumulativeHeldBtc of 1,329,077.2, with the previous point at -691.8 and 1,329,664.3. The arithmetic closes: 1,329,664.3 minus 587.1 is 1,329,077.2, which is the cumulative the next day reports. A flow series that reconciles internally is a series an agent can chain across days without a correction step.

Resources and prompts

resources/list returned 116 items: kote://catalog (every chart with formula, fields, input categories and limitations as JSON), kote://methodology (sources by category, units, cadence, limitations and citation policy), and 114 per-chart documents at kote://charts/{id}. resources/templates/list confirms the kote://charts/{id} template.

prompts/list returned two: explain_metric (takes a required id) and cycle_position_brief (no arguments), the latter described as a short sourced read of cycle position using MVRV Z-Score, NUPL, SOPR and exchange flows. Both are editable prompt surfaces a client can invoke without the operator writing the prompt.

Downsampling, argument names and the defects

Four things are worth knowing before integrating, and three of them are recorded because they were measured rather than read:

  • The required argument is id, not chart. A call using chart was refused by JSON-schema validation with a message naming id as required, which is the correct behaviour and a fast correction for a caller working from prose. The submission text and the vendor's marketing copy both speak about "the chart" rather than the argument name.
  • additionalProperties: false is declared but not enforced, and the downsample argument is maxPoints. Two calls passing points (6 and then 100000) were both accepted and silently dropped, and both returned the default 500 rows. maxPoints is documented 10 to 2000 with a default of 500. Anyone who reads the submission's downsampled to the points you ask for and passes points will get 500 rows every time and no error, so this is the one place where a caller can be wrong without being told.
  • The running instance is behind its own registry record. initialize reports kote-analytics 1.1.0 while the official registry lists 1.1.1 as the latest published version. Nothing measured here failed, but a version skew between the deployed server and its published record is worth knowing when a later call behaves differently from the registry entry.
  • Freshness is per chart, and it is reported honestly. Both get_latest answers were published at 2026-11-16T14:16Z from the same processing run at 13:53Z on block 976,137, yet the MVRV Z-Score ran through 2026-11-15 while spot ETF flows stopped at 2026-11-13. The asOf.lastT field carries the difference, so a caller that always reads lastT instead of assuming today will not overstate the freshness of the flow line.

The refusals are clean: get_latest on an unknown id answered with an error naming the bad id and telling the caller to run list_charts to discover valid ones, rather than returning an empty series that reads as a zero or a gap.

Installation

Claude Code:

claude mcp add --transport http kote-charts https://kotecharts.com/api/v1/mcp

Any client that takes a JSON config:

{"mcpServers": {"kote-charts": {"type": "streamable-http", "url": "https://kotecharts.com/api/v1/mcp"}}}

Local stdio bridge for a client that cannot reach a remote server:

npx -y kotecharts-mcp

If the operator would rather call REST directly, the same data is at https://kotecharts.com/api/v1/public/charts/{id} behind a free key from kotecharts.com/account at 3 requests a second, and the metric catalog is open at /api/v1/metrics/{id} with no key at all. Python (pip install kotecharts) and TypeScript (npm i kotecharts) clients are published beside it.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations

  • The data licence is not the code licence. The repository's code is MIT, but the data served by the site, the REST API and this MCP server is governed by the vendor's terms: free for personal, educational, research and other non-commercial purposes, with attribution, and commercial redistribution requires contacting the vendor. Each response carries a cite string such as Kote Charts - "MVRV Z-Score", https://kotecharts.com/c/mvrv-z-score, and the attribution is part of the licence rather than a courtesy.
  • One asset. This is Bitcoin on-chain analytics only. There is no Ethereum, no stablecoin supply beyond the Tether comparison chart, and no equity, ETF or macro series except the cross-asset comparisons the catalog names.
  • Heuristic estimates inside a precise-looking surface. The vendor states that metrics are heuristic estimates, entity labels in particular, and asks that the limitations be read before conclusions are drawn. Coverage of exchange and entity attribution depends on clustering heuristics that off-chain flows defeat by construction.
  • Rate limits are the free tier. 60 requests a minute per IP keyless, or 3 requests a second with a free key on REST. A loop over 114 charts, or a nightly sweep of every metric, will hit the keyless ceiling and needs pacing or a key.
  • The full-history call is downsampled by default. Asking for a chart with no range returned exactly 500 rows spanning 2010-08-18 to 2026-11-15, which is even downsampling and not the raw daily series. The catalog's minDate of 2009-01-03 is the data window, while the first populated MVRV row is 2010-08-18 because early price data supports values only from there. A caller that needs every day must page with from and to or raise maxPoints to its ceiling of 2000.

See Also

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing