Cursor Agent SDK vs TradingAgents

Cursor Agent SDK

An agent-as-code framework with typed tools, declared read and write effects that drive retries and dry runs, an approval gate per tool and evals that assert on tool calls. It's closed source, every model turn runs through Cursor's backend on a Cursor plan, local sandboxing is off by default, and the changelog has two entries.

TradingAgents

An Apache 2.0 LangGraph framework at 0.7.0 with six releases in 90 days, a dated changelog with upgrade notes, a public test matrix and point-in-time rules for historical runs. It has no MCP client or server, no PyPI package and no documentation beyond the README, and its output is a model-written rating that places no orders.

Cursor Agent SDK

1. Install @grok/agent-sdk, not @cursor/agent-sdk from the launch post. That name answers 404 on npm
2. Set local.sandbox before a turn reads webhook payloads, pull request diffs or chat. Without it shell commands auto-approve with the host's credentials
3. Give every tool an effect. A tool without one counts as a write, runs once and is stubbed in dry runs
4. Start agent-sdk serve with --bearer-token on any shared host. --allow-anonymous admits every caller as one principal
5. Delete bot/hooks/memory.ts when sessions serve different people. The journal shares each turn's message and result with later turns

TradingAgents

1. Run headless with tradingagents --ticker <T> --date <YYYY-MM-DD> --analysts market,news,fundamentals --save --no-show; a missing answer stops the run and names the flag
2. Guard on the REVIEW signal before mapping propagate()'s decision to a rating; it means no rating could be parsed
3. Set SEC_EDGAR_USER_AGENT to the operator's own name and address before fetching US fundamentals
4. Keep provider keys out of the working directory tree unless intended, since the package loads the nearest .env on import
5. Treat the rating as model output for research; the framework places no orders and its results vary between runs