12research roles
TradingAgents Report is a multi-agent research desk that turns one listing into evidence, debate, risk review, and a five-tier final rating. Research only. No trade execution.
New accounts receive free analysis credits
Evidence modules, Bull/Bear debate, planning, risk review, and the Portfolio Manager form a traceable research path.
The Portfolio Manager issues the product conclusion: Buy, Overweight, Hold, Underweight, or Sell.
US, Hong Kong, China A-shares, Japan, and Taiwan, subject to the deployment whitelist.
Practice with virtual cash and delayed quotes. No broker connection or live orders.
Reports
Reports the team explicitly publishes to this site. Each one is a multi-agent research reference with an analysis date and final Portfolio Manager rating. Not investment advice or trade execution.
Hold Shell Plc (SHEL.L) at 3,539 GBX with an active stop at 3,420 GBX; let LNG Canada Phase 2 FID (early Oct) and Q3 earnings (Oct 29) break the balanced risk/reward tie.
SHEL.LLSE
Trim HSBA.L to 50–75% of benchmark into the 1515–1540 shelf; Underweight with a 1456 GBX target, completing the reduction on a daily close below 1488.
HSBA.LLSE
Hold GLEN.L at benchmark weight — no add at 556.80, no exit; let price versus 584.47 and 525.94 plus FCF and litigation disclosure decide the next move.
GLEN.LLSE
Underweight BATS.L — trim roughly one-third to ~60-65% of benchmark on the broken medium-term trend, keeping a residual for the yield and buyback bid.
BATS.LLSE
Underweight RR..L: trim one-quarter to one-third into the 1468-1489 resistance cluster, keep a sub-half core, and re-add only on a volume-confirmed reclaim above 1489.
RR..LLSE
Research MCP
Cursor, VS Code, Claude Code, Codex, and Gemini can start analyses, read the watchlist, and place simulated orders. Add a URL, then sign in. Do not put tokens in the config file.
One click or one command
LSE
claude mcp add --transport http tradingagents-report \
https://mcp.tradingagentsreport.com/mcp
codex mcp add tradingagents-report --url \
https://mcp.tradingagentsreport.com/mcp
codex mcp login tradingagents-report
gemini mcp add --transport http tradingagents-report \
https://mcp.tradingagentsreport.com/mcpHow research forms
Optional specialist modules gather evidence. Bull and Bear debate, a Research Manager plan, Trader proposal, and risk review always follow. The Portfolio Manager issues the final five-tier rating.
Read the full research method→Choose Market, Sentiment, News, and Fundamentals modules. Selected analysts run in a fixed order from the confirmed listing and analysis date.
Bull and Bear researchers challenge the evidence. The Research Manager drafts an investment plan; the Trader proposes Buy, Hold, or Sell as an intermediate step, not the final product rating.
Aggressive, Conservative, and Neutral risk voices stress-test the plan. The Portfolio Manager decides Buy, Overweight, Hold, Underweight, or Sell.
What you receive
Read the Portfolio Manager decision first, then the Markdown from roles that ran.
The Portfolio Manager issues Buy, Overweight, Hold, Underweight, or Sell.
Conviction, reference price, position, zones, target, and key risks.
Market, sentiment, news, and fundamentals signals appear only when their modules ran.
Full specialist, debate, risk, and manager reports, exportable from the product app.
Simulated Trading
A five-tier rating is still a hypothesis. Simulated Trading records what you would buy or sell with 10,000 USD virtual cash, delayed quotes, and a server-side ledger. It is not a brokerage account.
Equity, cash, positions, and P&L persist on the server. Opening grant is 10,000 USD simulated cash. The browser is never the account of record.
Place a simulated buy or sell from a stock page. Market orders need an open session; limit orders can wait for the next tradable session.
Simulated cash is not used to pay fees. Fees settle in credits from notional, first from plan points then bonus points.
Attach a short note to the order. Review fills, T+1 sellability, and unrealized versus realized P&L in Simulated Trading.
Markets and data
Research runs only on admin-enabled exchanges after listing identity is resolved. Every report records an analysis date; quotes can be delayed.
Default seeds include US (NASDAQ, NYSE, AMEX, OTC), HKEX, China A-shares (SSE, SZSE), Japan (TSE), and Taiwan (TWSE, TPEX). Live availability follows the deployment whitelist.
Ticker, exchange, and asset identity are resolved before research. Provider-specific symbols are not the public instrument identity.
Prices, fundamentals, and news stay inside the selected analysis date. Missing source fields are not invented; social or prediction inputs may reflect live-at-run data.
Research desk
Sign in to search a listing, set the analysis date, choose evidence modules and models, then track the job through to a report.
Search by symbol, confirm the exchange, and keep ticker identity separate from vendor-only codes.
Market, fundamentals, news, and sentiment are optional. Debate, risk review, and the Portfolio Manager always run.
Submit up to five instruments at once. Each creates its own analysis job and credit check.
Pick quick-think and deep-think models from the admin-open catalog. Both stay on the same provider for one job.
Plan points and bonus points are tracked separately. Successful or user-cancelled runs settle from actual AI cost; system failures are not charged.
New accounts receive free analysis credits
Start free research→Apache 2.0 on GitHub
The multi-agent graph—evidence, debate, risk review, and a five-tier rating—is public. Clone it, bring your own keys, run the CLI. The hosted desk is the product around that engine: accounts, credits, and published reports.
Exchange-level symbols, bars, indicators, news, and fundamentals. Yahoo Finance remains a fallback.
Analyst signals, a decision brief, and in-window chart snapshots—not only Markdown.
Deterministic parsing for NASDAQ:AAPL, 600519.SS, and HKEX:700.
TTL cache by default. Optional Redis if several processes share a ticker.
git clone https://github.com/tradingagents-report/tradingagents-report.git
cd tradingagents-report
python3.12 -m venv .venv
source .venv/bin/activate
pip install .Python 3.12. Redis and PostgreSQL are not required for a local run.
CLI and Python entry point. You supply LLM and market-data credentials. Runs stay on your machine.
Sign-in, usage-point billing, private jobs, Simulated Trading, MCP, and curated public reports. Not in the clone.
Apache 2.0 derivative of Tauric Research / TradingAgents (arXiv:2412.20138).Read the paper explainer→
Usage capacity
Monthly plans grant cycle usage points for the research desk. New accounts can also start with signup bonus points. Live prices and entitlements are confirmed in the product billing page.
Entry cycle capacity for individual research workflows.
2,000 usage points / cycle
Higher monthly capacity for more frequent ticker coverage.
5,000 usage points / cycle
Expanded capacity for broader watchlists and small teams.
10,000 usage points / cycle
Highest published monthly capacity on the public catalog.
20,000 usage points / cycle
Marketing catalog only. Points are not one report per point: finished jobs settle from actual AI cost with the active markup. Checkout always uses the live catalog in the product app.
Clarity first
TradingAgents Report is a hosted multi-agent equity research desk built on the TradingAgents framework. Optional Market, Sentiment, News, and Fundamentals modules gather evidence; Bull/Bear debate, Research Manager, Trader, risk review, and Portfolio Manager always run to produce a traceable report and five-tier final rating. Live brokerage orders are not placed. Optional Simulated Trading is a separate virtual ledger.
No. Research reports do not connect to a brokerage, manage funds, or place live orders. Simulated Trading, when enabled, records market and limit orders against delayed quotes and 10,000 USD virtual cash. Fills are internal simulation results, not broker executions.
Each report records an explicit analysis date. Historical prices, fundamentals, and news are bounded to that date. Quotes disclose source and time and may be delayed. Some sentiment or prediction-market inputs can reflect live-at-run data rather than a historical as-of snapshot.
LLM behavior, temperature, data availability, and nondeterminism all affect output. The agent graph keeps roles and evidence flow consistent so you can compare process quality across providers such as OpenAI, Anthropic, Google Gemini, and DeepSeek when those models are enabled in the catalog.
Plan points refresh each billing cycle; bonus points from signup, referral, or admin grants can carry over. Available balance must be above the analysis threshold to start. Finished jobs settle from actual AI USD cost times the configured points-per-USD and markup. System failures are not charged; user cancel can still settle actual cost. Checkout always uses the live catalog in the product app.
No. Reports are informational research references, not personalized investment advice. Verify important evidence independently and decide within your own risk framework.
Simulated Trading is a practice ledger in the research desk. New accounts start with 10,000 USD virtual cash. You place market or limit orders from a stock page; fees settle in usage points, not simulated cash. Delayed quotes drive fills. There is no broker connection, and a report never places orders automatically.
Yes. The research engine is public at github.com/tradingagents-report/tradingagents-report under Apache License 2.0. You bring LLM and market-data keys. The hosted desk, billing, MCP, Simulated Trading, and public reports are the product—they are not in that clone.