Skip to main content
The Noonum MCP server exposes Noonum strategy and company analytics to MCP-compatible LLM clients (Claude Desktop, Cursor, and similar). It offers 17 tools that cover the strategy workflow of the REST API. Each tool calls the REST API with your own credentials, so the same ownership rules and plan credits apply as when you call the REST API directly.

Connecting

Point your MCP client at the base URL above over the MCP Streamable HTTP transport. No SSE or custom-protocol configuration is required. For a walkthrough of adding the server to Claude Code, Claude Desktop, Cursor, or a similar client, see MCP client setup.
The base URL is permanent and has no version in its path.

Authentication

Every request carries Authorization: Bearer <token>. The token is either a Noonum API key or an access token from the MCP server’s OAuth sign-in. An API key needs no OAuth round trip, so it suits server-side and CLI integrations. With OAuth, your client handles the sign-in through the OAuth 2.0 Authorization Code flow with PKCE. It discovers the MCP server’s OAuth metadata, registers itself, and opens the Noonum sign-in page in your browser. The client also receives a refresh token, so you don’t have to sign in again each time the access token expires. For a per-client walkthrough, see MCP client setup. For API keys, see Authentication. API-key authentication failures return a standard 401 Unauthorized; OAuth failures cause the client to re-initiate the authorization flow. Both are covered under Troubleshooting.

What your agent can access

Once connected, an agent acts as you and can:
  • Read your strategies, their ranked companies, conviction scores, evidence and reasoning, and factsheets.
  • Check your plan and the credits left in its current period.
  • Read public data (the public strategy catalog and company search) available to any authenticated user.
  • Create, update, archive, and revive your own strategies, and run them.
The Noonum MCP server is an analytics surface: it returns research and rankings. It does not place trades, move money, or connect to any brokerage account. There are no order, balance, or position tools, and none can be added through this server. Access falls into two tiers:
  • Private: your own strategies, including archived ones, their versions and runs, and your plan’s credits. Only you can update, archive, revive, or submit your strategies.
  • Public: the public catalog (list_strategies with scope set to public) and search_companies, open to every signed-in user. You can read a public strategy you don’t own and its results. You see only its active version, and its user_id is null.
Eleven tools are read-only. create_strategy and request_strategy_factsheet add data, and revive_strategy restores an archived strategy. update_strategy, archive_strategy, and submit_strategy are marked destructive. Every tool declares the standard MCP tool hints (readOnlyHint, destructiveHint, idempotentHint, and openWorldHint), so a client that honors them may ask you to confirm the destructive ones. Each new run uses one credit from your plan. A failed run is not charged.

Disclaimer

Noonum tools return informational research and analytics. This is not investment advice and not a recommendation to buy or sell any security. Output is not guaranteed to be accurate, complete, or suitable for your purposes. Verify it independently before relying on it, and own any decision you make from it.

Tools

An agent can, on your behalf:
  • Turn an investment objective into a ranked, scored list of companies.
  • Show the reasoning and evidence behind a company’s place in a strategy.
  • Filter rankings by exchange region, domicile, and market cap, and read them as of a past date.
  • Search for companies and browse the public strategy catalog.
  • Check the credits left on your plan.
The MCP server exposes 17 tools, grouped by task.

Strategies

  • list_strategies (scope?, limit?, offset?): list strategies one page at a time. scope is user (your own, the default), public (the public catalog), or all (both). limit is 1 to 50 (default 20). Each strategy carries its id, name, objective, public, status, completed, and summary performance metrics. total counts strategies across all pages.
  • get_strategy (strategyId): get a strategy and the run state of its active version, with summary performance metrics but no weekly performance series. user_id is null when you are not the owner.
  • create_strategy (name, objective?, exclusions?, waivers?, mode?, tags?, filter_intent?): create a strategy. Nothing runs until you submit it. filter_intent stores the default filters for get_strategy_companies (see Filters and metrics).
  • update_strategy (strategyId, plus at least one create_strategy field): update the draft. A field you send replaces its stored value, so exclusions and waivers must carry every phrase to keep. Changes take effect on the next submit.
  • archive_strategy (strategyId): archive a strategy. You can revive it for 30 days. After that, it is permanently deleted.
  • list_archived_strategies: list your archived strategies that are still inside the 30-day window, most recently archived first.
  • revive_strategy (strategyId): restore an archived strategy inside the 30-day window.

Runs and versions

  • submit_strategy (strategyId, as_of_date?, version_id?, force?, create_factsheet?, priority?): start a run and return the version with its run state. It does not wait for the run to finish. See Runs.
  • list_strategy_versions (strategyId): list a strategy’s versions, newest first, each with its latest run state.
  • get_strategy_version (strategyId, versionId, asOfDate?): get a version and its run state. Poll it after a submit.
  • get_strategy_available_dates (strategyId, versionId?): list the as-of dates that have a completed run.

Results

  • get_strategy_companies (strategyId, asOfDate?, versionId?, includeReasoning?, exchangeRegion?, domicile?, marketCapMin?, marketCapMax?, limit?, offset?): list the companies a run produced, one page at a time, highest convictionScore first. limit is 1 to 100 (default 50). includeReasoning adds the reasoning to each row and makes it much larger, so keep limit at 20 or less with it. total counts companies across all pages.
  • get_strategy_company (strategyId, companyId, asOfDate?, versionId?): get the reasoning and evidence behind one company’s inclusion or exclusion. isIncluded says which. Each item in summaries has a provider, a pubDate in ISO 8601 UTC (such as 2026-07-08T20:18:06+00:00), and the text.
  • get_strategy_factsheet (strategyId, asOfDate?, versionId?): get the factsheet PDF of a run. status is ready with a url to the PDF, generating, not_requested, or unavailable. The URL expires after about 5 minutes, so call the tool again for a fresh one.
  • request_strategy_factsheet (strategyId, as_of_date?, version_id?, force?): start generating the factsheet of a completed run. Only the strategy’s owner can call it. Only the first request starts a generation, so asking again starts nothing new. Set force to regenerate a factsheet. If a generation does not finish, the status goes back to not_requested and you can request it again. The tool returns 409 while the run for that date has not finished.

Search and credits

  • search_companies (query, limit?): search by name, ticker, ISIN, or FIGI to find a company ID. limit defaults to 10.
  • get_usage: get your plan, the credits used and left in the current period, and when they refill (refills_at). On an unlimited plan, granted and remaining are null.

Filters and metrics

get_strategy_companies applies the strategy’s stored filter_intent when you pass no filter parameter. If you pass any filter parameter, the stored filter_intent is ignored, so pass every filter you want:
  • exchangeRegion: region labels (such as United States, Europe, or Japan) or ISO 10383 MICs (such as XNAS). They pick which listing represents each company. A company with no matching listing is dropped.
  • domicile: region labels or ISO 3166-1 alpha-2 codes (such as US). A company domiciled elsewhere is dropped.
  • marketCapMin and marketCapMax: inclusive bounds in USD. A run that carries no market cap ignores the bounds.
[] for exchangeRegion or domicile means no filter on that dimension. To read a run without the stored filters, pass exchangeRegion as []. When a filter applies, marketBuzz and convictionScore are recomputed over the companies it keeps. The metric fields carry these meanings (see Signals and scores for the full glossary):
  • convictionScore: a score from 0 to 1 that combines linguisticBeta and marketBuzz.
  • linguisticBeta: the company’s exposure to the strategy theme.
  • marketBuzz: the company’s share of the market buzz across the companies returned.
Compare these values within one strategy only.

Runs

submit_strategy starts a run and returns right away with the version and its run state. The run goes through Noonum’s ranking and scoring pipeline. It does not place a trade or execute an order.
  1. Call submit_strategy. Omit as_of_date for a live run, or pass a past date for a historical run.
  2. Poll get_strategy_version with the returned version_id until completed is true. A live run usually takes a few minutes.
  3. Read the results with get_strategy_companies and get_strategy_company.
status reports progress: 1 to 95 while the run is in progress, 100 when it is done, -1 for a temporary error that is retried automatically, and -2 when the run failed. Each new run uses one credit from your plan. A failed run is not charged, so you can revise the inputs and submit again. If the version and date already have a completed run, a submit without force returns that run and uses no credit. Set force to run it again. as_of_date can go back to 2019-01-01. Runs dated before 2025-01-01 have no market caps and no factsheet. For historical runs, priority sets the queue order from 1 to 10 (default 5), and a higher value runs sooner. Below 5, runs can take hours. When the REST API rejects a call, the tool returns an error with the HTTP status code and the REST API’s message. A submit fails in these cases:
  • 429: your plan’s credits for the current period are used up. get_usage shows when they refill.
  • 423: a run for the same version and date is already in progress. Wait for it to finish.
  • 409: you signed in with OAuth, and your email already belongs to another Noonum account. Sign in the way you first signed up, or contact support. get_usage returns the same error.
  • 404: the strategy does not exist, or you are not its owner.
  • 400: an input is out of range or malformed, such as a future as_of_date. The message names the bad value.

REST API mapping

Each MCP tool calls one Noonum REST API endpoint. Paths are relative to https://api.noonum.ai/v2.

Notes & limitations

  • Not every REST endpoint has an MCP tool. Portfolios, portfolio optimization, backtest, exposures, the evidence list for all companies (GET /strategies/{strategyId}/evidences), the runs of a version by date (GET /strategies/{strategyId}/runs), and the objective helpers (such as POST /enhance-objective and POST /reverse-objective) are available through the REST API only.

See also