Skip to content
Meet HeadstartYour market research, in one place
MCP · AI integration

Your assistant.
Headstart’s research.

Connect a compatible AI assistant to Headstart’s read-only research tools. Ask in plain English, then inspect the information behind the answer.

↗Your AI assistant

You ask the question

MCPA read-only connection

Tools with defined permissions

HHeadstart research

Stored data, dates, and limitations

Production catalog

Start with definitions and proof counts.

The production tool catalog is limited to hs.explain_metric and hs.proof_summary, plus approved metadata resources and the market-research prompt.

Service availability still depends on deployment configuration. A credential does not unlock disabled tools.

Development / test catalog

19 tools. A broader research toolkit.

Includes market, company, options, earnings, portfolio, and watchlist research, with 13 resources/templates and 3 prompts.

Provider-backed production access remains restricted pending written redistribution approval.

From Account → AI Connections

Connect without handing over control.

  1. Create a named credential.

    Open Account → AI Connections. Give the connection a name you’ll recognize, such as “Research assistant.”

  2. Choose permissions and expiry.

    Use full read-only access or choose individual scopes. Available lifetimes are 30, 90, 180, or 365 days.

  3. Save the one-time secret securely.

    The hs_mcp_* secret is shown once. Store it in your client’s secret manager—not a public config, screenshot, or chat.

  4. Configure and check your client.

    Use the MCP endpoint supplied by your operator, or a trusted local checkout. Refresh tool discovery and start with “Explain Quant Score.”

Open AI Connections →
You choose the access

Read-only, scoped, revocable.

  • Market context · screens · ticker research
  • Options flow · events · track record
  • Model portfolios · portfolio analysis
  • Revoke an active credential from your account

Access also depends on your account entitlement and which capabilities are enabled. The connection cannot place trades, change your watchlist, or refresh data providers.

The current account UI labels this section “AI Connections.” This guide uses the same credential options.

Client configuration

Two supported connection paths.

These are configuration templates, not live credentials. Your MCP endpoint comes from your operator.

Remote · Streamable HTTP
{
  "mcpServers": {
    "headstart": {
      "type": "remote",
      "url": "<HEADSTART_MCP_URL>",
      "headers": {
        "Authorization": "Bearer <secret-store-reference>"
      }
    }
  }
}

Replace both bracketed values using your deployment details and client’s secret-storage mechanism. Client configuration syntax can vary.

Local checkout · stdio
{
  "mcpServers": {
    "headstart": {
      "command": "python3.11",
      "args": [
        "-m",
        "mcp_server",
        "--transport",
        "stdio"
      ],
      "env": {
        "MCP_ENABLED": "true",
        "HEADSTART_AGENT_API_URL": "http://127.0.0.1:8060",
        "HEADSTART_MCP_TOKEN": "<secret-store-reference>"
      }
    }
  }
}

Requires the installed Python 3.11 MCP service, its dependencies, a trusted checkout working directory, and a configured local backend. This page does not start a server.

Actual feature reference

Find the tool for your question.

Full development/test catalog. Production availability is narrower, as listed above.

19 tools in the development / test catalog

Market & discovery / 5 tools
  • hs.market_brief

    Stored momentum cards, market exposure, macro, sector regime, and cached record-volume context

    hs.market.read

  • hs.screen

    Curated screen catalog and bounded screen members, including platform pick families

    hs.screens.read

  • hs.ticker_search

    Local ticker/company lookup

    hs.ticker.read

  • hs.sector_regime

    Market-wide sector breadth, leaders, laggards, and regime labels

    hs.market.read

  • hs.record_volume

    Cached scheduled record-volume breakthroughs

    hs.market.read

Company research / 4 tools
  • hs.ticker_snapshot

    Stored price, Quant Score, explanation evidence, and crowd sentiment

    hs.ticker.read

  • hs.ticker_research

    Ticker technicals, flow, ownership/events, earnings, setups, and persisted IV context

    hs.ticker.read

  • hs.setup_flags

    Stored-price NR7 and inside-day setup evidence for one ticker

    hs.ticker.read

  • hs.iv_surface

    Persisted ATM IV, IV rank, term slope, realized-volatility context, method, and source

    hs.ticker.read

Options, events & earnings / 4 tools
  • hs.flow_intelligence

    Alpha/whale options-flow evidence with contract details and limitations

    hs.flow.read

  • hs.event_intelligence

    Stored insider, Congress, and sentiment events

    hs.events.read

  • hs.earnings_intelligence

    Stored earnings dates, expected moves, IV-crush estimates, and historical context

    hs.events.read

  • hs.dark_pool

    Disclosed deep-ITM options-flow proxy; not confirmed dark-pool activity

    hs.flow.read

Portfolios & watchlists / 3 tools
  • hs.model_portfolios

    Stored hypothetical model-portfolio summaries and holdings

    hs.portfolio.read

  • hs.portfolio_xray

    Stateless analysis of user-supplied holdings

    hs.portfolio.analyze

  • hs.watchlist

    The authenticated user's watchlist, read-only

    hs.portfolio.read

Track record & definitions / 3 tools
  • hs.proof_summary

    Rights-approved aggregate proof counts

    Public capability

  • hs.track_record

    Stored outcome statistics and provenance split

    hs.proof.read

  • hs.explain_metric

    Product-owned metric definitions, units, and limitations

    Public capability

hs.dark_pool is a deep-in-the-money options-flow approximation—not confirmed dark-pool activity. Watchlist access is read-only; portfolio X-ray analyzes the holdings you supply without saving changes.

Beyond individual tools

Resources to read. Prompts to start with.

13 resources and templates

Structured reference material and stored research. Availability follows the environment and permissions.

headstart://manifestheadstart://metricsheadstart://metrics/{key}headstart://screensheadstart://screens/{slug}headstart://screens/{slug}/historyheadstart://tickers/{symbol}/researchheadstart://flow/alpha-callsheadstart://eventsheadstart://track-record/{track}headstart://model-portfoliosheadstart://model-portfolios/{slug}headstart://model-portfolios/{slug}/backtest

3 research prompts

Try a real company: NVIDIA (NVDA). Its August 27, 2025 release reported $46.74B in quarterly revenue. Ask your assistant to compare the stored research with the company’s own report.

Company release ↗
hs.research.markeths.research.tickerhs.research.portfolio

Prompt examples for an enabled environment:

  • “Explain what Quant Score measures.”
  • “Summarize the stored market context and its dates.”
  • “Research NVIDIA (NVDA) and explain the dates and sources behind the answer.”

Use the company-research prompt only in an environment with the corresponding tools enabled. These are real tool use cases, not pre-generated AI answers. A prompt cannot bypass permissions.

Before your first question

A few practical answers.

Will this trade or update my account?

No. The MCP service is read-only. It does not execute strategies, place orders, mutate account data, or run provider refreshes.

Why can’t my assistant see all 19 tools?

The catalog is filtered by environment and enabled capabilities. Production currently permits only the two public tools described above. Permissions and account entitlement also apply to credentialed reads.

How should the assistant handle missing data?

Responses use the agent.v1 schema and explicit states: ok, partial, absent, computing, or error. Keep missing values and limitations visible; do not convert absence into zero or invent a result.

What if I lose or expose the secret?

Create a new credential and update your client. Revoke the old active credential in AI Connections. Never paste the secret into a public page or an AI conversation.

Can I use the same raw credential on data routes?

No. The raw AI credential is exchanged through the narrow introspection boundary for a short-lived delegated assertion. It is not forwarded directly to Agent-v1 data routes. Use the MCP client/service connection path.

Connect the assistant. Keep the judgment.

Manage the credential and permissions from your account.

Open AI Connections →