Give your agent
market history to reason with.
Connect your agent to market memory.
Add this URL to a compatible Streamable HTTP MCP client, with no authentication:
https://chartlibrary.io/mcpThe agent can answer that with market_state alone. Choose additional tools when the question needs published research. Connection guide →
Research tools. Start with your question.
Choose the question your workflow needs to answer. These examples show what each tool provides.
An example question
What does AAPL’s latest completed session resemble in history?
market_state(symbol="AAPL")What the tool returns
- Current and previous market-state context
- Historical comparisons and the range of outcomes that followed
- Earlier comparable state transitions across stocks
- Session dates, sample sizes, warnings and provenance
These ranges describe the historical sample. They are not automatically calibrated forecasts.
Illustrative call. This example does not run a live request.
An example question
What does the published daily research show, and how were its examples selected?
daily_note()What the tool returns
- The published daily research and its date
- The rule used to select the observations
- The daily research’s own settled-note tally
- Unsettled observations and limitations
Keep the publication date and selection rule. Outcomes that have not settled remain unknown.
Illustrative call. This example does not run a live request.
An example question
Which method does the published calibration evidence actually evaluate?
research_quality()What the tool returns
- The dated calibration receipt
- The specific method and population evaluated
- Sample sizes and qualifications
- Limits on applying the finding to other outputs
A finding applies to the method and sample it names. It does not validate every comparison.
Illustrative call. This example does not run a live request.
| MCP tool / REST request | What you get |
|---|---|
get_company_fileGET /api/v1/research-desk/company/FLNC | The same company file as the website: sourced financial history, thesis and operating scenarios where available. |
get_company_statusGET /api/v1/research-desk/company/FLNC/status | Review and source-scan clocks, pending evidence and incomplete checks. Read alongside the company file. |
get_company_revisionsGET /api/v1/research-desk/company/CIFR/history | Browse immutable revisions, publication clocks, reasons and pinned downloads. |
compare_company_revisionsGET /api/v1/research-desk/company/CIFR/changes?from_revision=3 | Compare facts, assumptions, financial history, scenarios and thesis between revisions. Omit to_revision for latest. |
run_company_scenarioGET /r/CIFR/lab | Open the multi-input model lab, or use the same read-only scenario API. Read the model first; overrides is a JSON object of 1-20 assumption cell IDs and values in published units. Calculate all dependent scenarios without saving or changing facts. Preserve model limitations. |
check_company_forecastsGET /api/v1/research-desk/company/CIFR/accountability?forecast_revision=3 | Compare frozen revenue forecasts with reported results. Forecasts must precede quarter end. Missing actuals, restatements and definition limits remain explicit; no calibration claim. |
compare_company_historyGET /api/v1/research-desk/compare?symbols=CIFR,WULF&period_end=2026-06-30 | Compare 2-6 companies (comma-separated tickers) at one reported quarter end in USD. Mismatched source concepts or missing history remain unavailable. Inspect provenance and fiscal-period differences. |
get_company_watchlistGET /api/v1/research-desk/watchlist?symbols=JBHT,VG | Latest published revisions, current review status, checkpoints and pinned downloads for 1-20 companies. Read-only; no server-side saving, new source review or live quotes. |
list_coverageGET /api/v1/research-desk/coverage | Browse companies and see which files are still skeletons. |
search_researchGET /api/v1/research/search?query=IONQ | Find published studies, articles and clearly labeled agent protocols. |
read_researchGET /api/v1/research/read?research_id=casebook:ionq-noon | Read findings, source documents, dates, sample receipts and limitations. |
market_stateGET /api/v1/state-packet?symbol=AAPL | One stock’s state, historical analogs, outcome ranges, and transition memory. |
daily_noteGET /api/v1/daily | The published daily research, its selection rule, and settled-note tally. |
research_qualityGET /api/v1/calibration | The dated calibration receipt, sample, and qualifications. |
get_company_file takes a ticker and optional as_of date; list_coverage needs no arguments. Company models label reported facts, guidance and desk assumptions separately. Browse the same company files. market_state takes a symbol and optional date. daily_note takes an optional date; its REST equivalent calls that parameter session. research_quality needs no arguments.
Start with a symbol.
curl "https://chartlibrary.io/api/v1/state-packet?symbol=AAPL"That is the complete request. It returns the latest built, completed-session state for AAPL. Add &date=YYYY-MM-DD for a specific historical session. No preceding search, comparison-set handle, or timeframe choice is required.
Latest means the latest available research session, not a real-time quote. Read the date in the response; a missing session remains missing.
Connect a market state to published research.
Find a relevant study, read its evidence, and cite the conclusion with its original scope.
Keep the qualifications with the number.
- Keep dates and counts. Preserve the returned session, analog count, distinct symbols and sessions, and each horizon’s observed sample size.
- Read the evidence status. Keep informative receipts, warnings, missing values, and provenance. An empty or weak result is not a directional signal.
- Know the units. State-packet outcome ranges describe historical excess returns in percentage points relative to a date-matched liquid-stock baseline.
- Use the right audit. Raw historical percentiles are not automatically calibrated forecasts. The calibration receipt applies to its named method and population; daily research has its own settled-note tally.
Historical comparisons are research evidence, not recommendations to buy or sell.
Exploring is free. Building something?
If you want to build a research workflow, integrate the memory into a product, or work with a larger study, talk to Graham. We can discuss the right interface, coverage, service requirements, and data permissions for your project.
Contact GrahamThe public tools retrieve market history and research. They do not store an agent’s private conversation history. Existing integrations keep their API routes and MCP tool names; the getting-started menu does not expose every experimental research method.
Research access has no paid tiers. Service protections still apply: honor rate-limit responses and retry guidance. Review data and licensing, terms, and privacy before redistributing outputs.