My Trading Agent
Checking…https://www.myfinance5051.com/api
MCP guide

System reference

Analysis architecture

One analysis engine serves the dashboard and MCP-connected assistants. The same run ID connects progress, charts, decisions and reports end to end.

The connections

Trader or assistant

Dashboard form or MCP client starts a run and checks its progress.

Frontend gateway

The dashboard reads the configured engine address. The MCP gateway exposes discoverable tools and checks a private access key.

TradingAgents engine

The API queues work, gathers market evidence, coordinates agents, and serves results and artifacts.

The hosted engine base is https://www.myfinance5051.com/api. Its direct MCP endpoint is https://www.myfinance5051.com/api/mcp/; the frontend MCP gateway is https://www.myfinance5051.com/mcp. The separate gateway and engine may enforce different access policies.

End-to-end analysis flow

  1. 01

    Submit

    A trader chooses a stock, date, depth, language, model and participating analysts. An assistant instead discovers analyze_ticker via MCP tools/list, then calls it with validated inputs. Both paths create a run in the same engine queue.

  2. 02

    Gather evidence

    The engine retrieves company and market data. Price series and technical indicators can become available before the slower AI reports; the dashboard polls run status and loads chart artifacts separately.

  3. 03

    Specialist research

    Selected market, fundamentals, news and sentiment analysts produce independent reports. Their output is provisional while the run continues; report count and model choice affect elapsed time and model costs.

  4. 04

    Debate and decision

    Bull and bear researchers weigh the evidence, a trader proposes a position, risk reviewers challenge it and a portfolio manager forms the final action and confidence assessment. Stages not selected by the trader can be omitted.

  5. 05

    Publish results

    The engine stores progress, a decision, individual reports and final result.html. It also publishes price-chart.json for interactive exploration and price-chart.png for a static report image. History lets both dashboard and assistant revisit runs.

Engine API and artifacts

Paths below are relative to the configured engine base. MCP uses JSON-RPC methods such as initialize, tools/list and tools/call rather than a separate URL for every tool.

GET /healthConnection status
POST /runsQueue an analysis
GET /runs/{id}Run and decision
GET /runs/{id}/statusLightweight progress
GET /runs/{id}/result.htmlComplete report
GET /runs/{id}/price-chart.jsonInteractive chart data
GET /runs/{id}/price-chart.pngReport image
GET /market/security?ticker=0700.HKHKEX stock names
POST /mcpAssistant protocol gateway

Access and limits

The frontend MCP tools require a private access key. Inputs are bounded; assistant-created runs are rate-limited, and cancellation through MCP is limited to MCP-owned runs. An analysis can use paid model credits. Keep access keys out of browser configuration and share them only with trusted clients.

Client configuration