@aihubspot/agent-trade-mcp 0.1.10 → 0.1.12

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +9 -7
  2. package/dist/index.js +521 -214
  3. package/package.json +1 -1
package/README.md CHANGED
@@ -1,6 +1,6 @@
1
1
  # AI Hub Agent Trade MCP
2
2
 
3
- Install with `npm install -g @aihubspot/agent-trade-mcp`, then start `ai-hub-trade-mcp --profile default`.
3
+ Install with `npm install -g @aihubspot/agent-trade-mcp`, then start `ai-hub-trade-mcp --profile default`. The default Toolset exposes the focused market, spot-order, universal-transfer, and wallet-query workflows. Use `--toolset full` to expose every supported Core capability. Read responses use compact text by default; use `--response-mode compat` only for clients that require the legacy duplicated JSON text.
4
4
 
5
5
  This package provides a local stdio MCP server. State-changing operations require a preview and a new explicit user confirmation.
6
6
 
@@ -15,6 +15,8 @@ ai-hub-trade-mcp setup --client cursor --profile default
15
15
  ai-hub-trade-mcp setup --client claude-desktop --profile default
16
16
  ai-hub-trade-mcp setup --client claude-code --profile default
17
17
  ai-hub-trade-mcp setup --client codex --profile default
18
+ ai-hub-trade-mcp setup --client codex --profile default --toolset full
19
+ ai-hub-trade-mcp setup --client codex --profile default --response-mode compat
18
20
  ```
19
21
 
20
22
  Setup registers the currently installed MCP binary through its absolute Node runtime and entrypoint path, so desktop clients do not depend on a global PATH or an unpublished `npx` package. Cursor and Claude Desktop configurations are merged with existing MCP servers through their JSON configurations. Claude Code and Codex are registered through their official CLIs; Claude Code uses user scope. Existing JSON configurations are validated, atomically updated, and backed up before their first modification. The setup command stores no API credentials; the MCP server continues to read its profile from `~/.ai-hub/config.toml`.
@@ -33,7 +35,7 @@ All read tools return `{ "ok": true, "data": ... }` in one of these stable forms
33
35
  - Objects: `{ "dataType": "object", "value": { ... } }`
34
36
  - Scalars: `{ "dataType": "scalar", "value": "..." }`
35
37
 
36
- MCP clients that support the current protocol also receive the same envelope in `structuredContent`, validated by each read tool's output schema. Check `data.dataType` before formatting; do not infer raw OpenAPI nesting.
38
+ MCP clients receive the full envelope in `structuredContent`, validated by each read tool's output schema. In the default `compact` mode, text contains only the response type plus safe list metadata (`count`, `returnedCount`, `totalCount`, `nextOffset`, or `truncated` when applicable), so the same JSON is not duplicated in the Agent context. Use `--response-mode compat` to also place the full envelope in text. Check `data.dataType` before formatting; do not infer raw OpenAPI nesting.
37
39
 
38
40
  ## Bounded market analysis
39
41
 
@@ -41,11 +43,11 @@ Use these tools for requests that would otherwise return a broad market payload:
41
43
 
42
44
  - `market_get_symbol_overview`: counts by quote asset and a small sample. Use this first for a generic request to list trading pairs.
43
45
  - `market_list_symbols`: a paged, optionally quote-asset-filtered list without order-rule metadata.
44
- - `market_search_symbols`: required-keyword lookup with at most 20 basic matching rows. Do not use it for a generic pair list.
46
+ - `market_search_symbols`: required-keyword lookup with at most 50 basic matching rows. Do not use it for a generic pair list.
45
47
  - `market_get_symbol_info`: exact precision and minimum order rules for one symbol only.
46
- - `market_get_ticker_summary`: watchlist, gainers, losers, and quote-volume leaders.
47
- - `market_get_depth_summary`: best bid/ask, spread, and up to 20 levels per side.
48
+ - `market_get_ticker_summary`: watchlist, gainers, losers, and quote-volume leaders under one total result budget (20 by default, 50 maximum).
49
+ - `market_get_depth_summary`: best bid/ask, spread, and up to 50 levels per side.
48
50
  - `market_get_trades_summary`: price range, buy/sell statistics, and up to 50 recent trades.
49
- - `market_get_klines_summary`: period change, high/low, latest candle, and up to 100 candles. It defaults to `60min`; formal intervals are `1min`, `5min`, `15min`, `30min`, `60min`, `1day`, `1week`, and `1month`.
51
+ - `market_get_klines_summary`: period change, high/low, latest candle, and up to 50 candles. It defaults to `60min`; formal intervals are `1min`, `5min`, `15min`, `30min`, `60min`, `1day`, `1week`, and `1month`.
50
52
 
51
- The complete raw symbols endpoint is intentionally CLI-only because its payload is too large for Agent contexts. Raw ticker, depth, trades, and kline tools remain available only when an exact payload is explicitly needed. Raw trades are capped at 100 rows; raw klines support the upstream maximum of 300 rows.
53
+ The default Toolset exposes only these bounded forms. The `full` Toolset additionally exposes raw symbols, depth, trades, and kline responses for explicit advanced use. Every list is limited to 50 rows; raw symbols are paged with `offset` and `nextOffset`.