Insights
Connect Claude to Structured Stock-Market Data
Alphanume Team · August 26, 2026
Use one hosted OAuth connector to give Claude structured, queryable stock-market data without passing CSV extracts or maintaining custom API glue.
Claude can query Alphanume's stock-market datasets directly after you paste one URL and sign in. The connector presents 25 read-only tools with named filters and structured responses. That is materially different from attaching a CSV: Claude can request the narrow slice needed for the question, preserve the response fields, and call a second dataset without waiting for another upload.
It is also different from asking Claude to write a temporary API script. The hosted MCP server already maps each tool to the corresponding Alphanume REST endpoint. Both paths return the same deterministic point-in-time records. MCP changes how the request is assembled and delivered, not the underlying dataset or its entitlement rules.
Connect the hosted server
In Claude web or desktop, open Settings, select Connectors, choose Add custom connector, and paste the server URL:
https://mcp.alphanume.com/mcpChoose Connect and complete the browser sign-in with the Alphanume account used for the subscription. The connector then carries across Claude web, desktop, mobile, Cowork, and Claude Code for the signed-in Claude account. Open /mcp in Claude Code to confirm it appears there. The MCP server documentation has the current endpoint and authentication reference.
Why a connector is different from a file upload
Workflow | What Claude receives | Main failure mode |
|---|---|---|
CSV attachment | One static extract selected before the question is complete | Stale rows, missing columns, or an undocumented export filter |
One-off API script | Whatever the generated code requests and transforms | Hidden defaults, credential handling, or an incorrect endpoint parameter |
Hosted MCP connector | Structured tool responses from the same Alphanume REST data | Wrong tool choice, date filter, tier window, or interpretation |
A connector removes repetitive plumbing, but it does not remove research discipline. The model still needs a precise question, an information cutoff, and an output contract. Requiring the raw source date and exact tool filters makes a conversational query inspectable later.
Ask for the latest settled risk regime
The S&P 500 Risk Regime dataset is a compact connection test because each row contains only a date and a binary risk_regime value. A value of 1 means risk-off and 0 means risk-on. Historical labels remain fixed once published. Use this prompt after the connector is active:
Use only the Alphanume MCP server.
1. Call check_api_status.
2. Inspect the get_sp500_risk_regime tool definition.
3. Query the latest observation available to my account.
4. Return the raw date and risk_regime value.
5. Translate 1 as Risk-Off and 0 as Risk-On.
6. State the exact filters used and whether access limits affected the result.
7. Do not turn the label into a trade recommendation.The Risk Regime documentation states that results are ordered by date descending, so the first returned row is the latest one available under the account's tier. The dataset updates daily and is designed as an overlay for filtering, sizing, and regime-aware analysis. It is not a price target or a directional forecast for an individual stock.
Keep the response contract visible
Element | Expected form | Why retain it |
|---|---|---|
count | Number of rows in the response | Distinguishes a populated result from an empty one |
date | YYYY-MM-DD | Defines when the published regime applies |
risk_regime | 0 or 1 | Preserves the source value before adding a human label |
filters | Exact arguments used by Claude | Makes the request reproducible in MCP or REST |
Do not ask Claude to replace an empty result with a nearby date silently. The nearest available record may belong to another session or sit outside the intended cutoff. A useful failure response says whether the server was reachable, whether the account was authorized, and whether the data call returned empty or restricted.
Account for tier and interpretation limits
Free access covers the trailing 20 trading sessions with a one-trading-session delay. Pro includes current observations and full history. The OAuth connector uses the key and tier already attached to the signed-in Alphanume account, so reconnecting does not expand coverage or change a rate limit. Requests outside the Free window return a tier restriction rather than older rows.
- The binary regime compresses market conditions into one overlay and cannot explain every source of portfolio risk.
- Historical labels being fixed does not guarantee that a strategy conditioned on them survives costs or a new market regime.
- Claude can summarize a field incorrectly, so preserve the raw integer beside the translated label.
- The MCP server exposes deterministic data, but the model does not independently validate a trading edge.
Create a reproducible next step
Save the prompt and its returned date, integer label, and filters. On the next session, run the same prompt and compare the raw values before adding more datasets. If you need a longer regime study, review access on the pricing page and request an explicit date range rather than "all history." For a wider view of agent-assisted research controls, continue with Using Claude for Trading Research.