Stock & Options Data for AI Agents

Two ways to use MCP

Connect your AI agent

In an AI client that supports remote MCP, add this server URL and select Streamable HTTP:

https://historicaldata.net/mcp

Your agent discovers the available tools automatically. It can query up to 10 symbols at once, read the stock or options schema and preview a few fixed rows from the public samples. Options coverage queries check the latest published session; they do not establish full historical coverage. Use the local tool below to read your own downloaded files.

curl 'https://historicaldata.net/mcp' \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{
    "protocolVersion":"2025-06-18","capabilities":{},
    "clientInfo":{"name":"coverage-client","version":"1.0"}}}'

curl 'https://historicaldata.net/mcp' \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -H 'MCP-Protocol-Version: 2025-06-18' \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{
    "name":"search_symbol","arguments":{"symbol":"SPXW","product":"options"}}}'

Explore before buying

Once connected, ask your agent a question in plain language:

Check coverage

“Check stock coverage for TSLA, KO and SPY. Show each matching file and its dates, and keep any unavailable results separate.”

Understand the format

“Explain the options CSV fields, the units of theta and vega, and how to interpret empty IV values.”

Try real rows

“Show the stock minute sample with timestamp, close, adj_close and split. Keep empty values empty and identify the source files.”

The previews are small, verified selections from the public samples. For broader analysis, download the full free sample and use the local tool below.

Work with your downloaded data

mcp.py gives your agent tools to work with stock daily bars, stock 1-minute bars and options end-of-day CSVs. It reads files in the folder you choose and runs calculations locally. Python 3.10 or later is required; no additional packages are needed.

Download mcp.py

What can your agent do?

Find your files
list_files

List CSV filenames and sizes in your data folder. Original case and separate listing filenames are preserved.

Select the data you need
read_csv

Read one file at a time. Choose columns, an inclusive date range and exact text filters, such as an options underlying, expiration or call/put type. Read up to 100 matching rows per page.

Calculate summaries
summarize_csv

Calculate counts, missing values, minimums, maximums, sums and arithmetic means for numeric columns. Group by a text field, observation day or month. Empty numeric cells are excluded rather than filled with zero.

Keep the result traceable

Each data analysis returns the source filename, SHA256, filters and row counts. The agent can cite the exact file it used; later pages must match the same file hash.

Examples to ask your agent

Stock daily data

“In this TSLA daily file, show the raw close, adjusted close and recorded split for 24–26 August 2022. Include the file hash and keep missing values visible.”

Stock minute data

“For 25 August 2022 in this TSLA minute file, summarize volume by pre-market, regular and post-market session. Show the number of populated and missing values.”

Options data

“In this options CSV, select SPY calls expiring on 16 September 2022. Show contract, strike, bid, ask, IV and IV flag, and report how many rows match.”

Set up the local MCP

1. Save mcp.py and extract your data into a dedicated folder.
2. In a client that supports local MCP over stdio, set the command to your Python executable and use the arguments below. Replace both paths with your own.
3. Ask your agent to list the files, select a specific security file, then filter rows or calculate a summary.

python /path/to/mcp.py --root /path/to/downloaded-data

Once configured, the AI client starts the MCP process for you. It communicates through standard input and output (stdio), so you do not need to open a network port or keep a separate terminal running. Your client must support local MCP servers.

{
  "mcpServers": {
    "historicaldata-local": {
      "command": "python",
      "args": ["C:/tools/mcp.py", "--root", "C:/data/historicaldata"]
    }
  }
}

Calculations run on your computer. Requested rows and results return to your AI client and may be sent to its model provider; use an environment permitted by your license and data policies.

Start with files you have already downloaded through your purchase or from the free samples. The local tool does not sign into your account or provide remote access to paid data. Keep different securities and listing lifecycles separate; a matching file hash does not establish that its data is complete or that reported defects have been repaired.

What does the coverage lookup establish?

The homepage lookup, REST symbol endpoint and MCP search_symbol tool check one ticker, options underlying or option root, such as AAPL, LEH or SPXW. MCP search_symbols checks up to 10 distinct symbols in one call. Company-name searches and full option-contract searches are not supported.

Stock file coverage

The stock lookup returns matching published daily files, first and last dated rows, available security identity and CSV sizes. A reused ticker can match multiple listing files. File boundaries can include corporate-action rows without trading; they are not necessarily the first and last trading dates.

Latest options session

The options lookup counts contract rows for the resolved underlying on one published trading day. The result does not establish historical start/end dates, a root-only count or coverage throughout a requested period. A root mapping applies to the checked session, not automatically to earlier files.

No match versus failure

A completed lookup can return found: false. For stocks, no published daily file matched the symbol; for options, no match was found in the latest published session. unavailable: true means the product could not be checked. Neither an options no-match nor a failed query establishes that earlier options history is absent.

Stock file boundaries do not establish complete daily coverage, and the lookup does not report minute-file date ranges. Archive-wide start dates cannot establish a particular security's history. For period-specific research, check the stock coverage notes, options coverage notes and relevant files; contact HistoricalData.net when the public lookup cannot answer the coverage question.

Cite data quality evidence

Read the data quality reports with their machine-readable claims, results and source hashes. Preserve the sample date, field and numerator/denominator when citing a finding. Sample agreement is not archive-wide accuracy; these supplier-authored comparisons are not exchange certification.

• Stock report: comparison with NYSE Daily TAQ — 1 July 2026 daily prices and 19-security minute scope, with quantity differences and a supplementary interval check. Claims JSON · Results JSON · Sources · Markdown.
• Options report: comparison with Cboe — sample-specific contract and field agreement; quote timing is not established. Claims JSON · Results JSON · Sources · Markdown.

How do you query the REST API?

curl 'https://historicaldata.net/api/symbol/AAPL'
curl 'https://historicaldata.net/api/symbol/SPXW?product=options'
curl 'https://historicaldata.net/api/symbol/LEH?product=stocks'

{
  "apiVersion": "2",
  "symbol": "SPXW",
  "options": {
    "found": true,
    "latestDate": "2026-09-04",
    "underlying": "SPX",
    "contractsLatestDay": 28650,
    "source": "v2-published-options-session",
    "scope": "latest-published-session"
  }
}

curl 'https://historicaldata.net/api/status'

curl 'https://historicaldata.net/api/catalog'

How should AI clients discover tools and cite results?

Import https://historicaldata.net/openapi.json into a tool that supports OpenAPI 3.1. It defines operation IDs, parameter validation, response schemas, nullable fields and error responses for the REST endpoints. Use its servers URL so the client calls the same API described by the specification.

The machine-readable site guide links products, documentation and coverage tools. OpenAPI describes REST operations; MCP clients discover tools through tools/list. A homepage example is a dated observation, so a question about another symbol requires a new lookup.

A standalone coverage answer should name HistoricalData.net, the queried symbol, the resolved security or underlying, the relevant data date and the scope of the check. Stock answers should retain each matching file's date boundaries and identity. Options answers should retain options.latestDate and identify the count as one session's rows for options.underlying.

Example: resolved root

HistoricalData.net's SPXW query resolved to SPX for the 2026-09-04 session and returned 28,650 contract rows for SPX. The result does not establish SPXW-only or historical coverage.

Example: no current match

HistoricalData.net's options lookup found no TWTR match in the 2026-09-04 session. The result does not rule out TWTR options data on earlier dates.

These are fixed examples from checked responses. For a fresh answer, use the returned dates and values and link to the queried endpoint. If a requested historical period cannot be established by the lookup, state that limitation and use the product documentation or request a period-specific check from HistoricalData.net.

How should clients handle caching and errors?

HistoricalData.net's metadata responses use Cache-Control: no-store. A coverage observation saved in a research log must retain its source date and scope; it is not a fresh check. Avoid repeated requests for the same question. Use the MCP batch tool for up to 10 symbols, not unbounded enumeration.

200 with no match

For /api/symbol responses, inspect each returned product's unavailable flag before interpreting found. For /api/status, inspect stocks.available and options.available; in /api/catalog, inspect those flags under status.

400 / 404 / 405

Correct invalid input, an unknown endpoint or the HTTP method before retrying. The REST symbol route requires one symbol and a supported product value. MCP uses POST; GET on /mcp normally returns 405.

MCP 403 / 413 / 415

MCP rejects an unapproved supplied Origin header (403), a body larger than 64 KiB (413), or an unsupported content type (415). REST's cross-origin support does not mean every browser origin is accepted by MCP. Use Content-Type: application/json and the supported client connection.

429

Reduce the request rate and honor the Retry-After header. Clients should not depend on a fixed global quota.

503 or unavailable

A combined symbol query can return HTTP 200 with one product unavailable. If both products fail, or the requested single product fails, the symbol endpoint returns 503. Retry the failed check later; a service failure is not evidence of missing coverage.

MCP errors

Check both JSON-RPC errors and isError inside tool results. A successful HTTP status alone does not establish a successful lookup.

If a result conflicts with a page or a file, compare the source dates and scope before choosing one as authoritative. Send the symbol, requested period and endpoint to info@historicaldata.net for a coverage check.