Introduction¶
DeepSeek Harness (DSH) adopts a plugin architecture designed to extend agent capabilities through plugins. However, integrating external MCP (Model Context Protocol) services usually requires manually writing configuration files, which adds extra work for developers.
dsh-algovault is a preconfigured DeepSeek Harness plugin maintained by AlgoVaultLabs. It mounts the AlgoVault MCP server into the Harness environment with a single command, allowing agents to directly invoke tools for composite trading signals, market regime detection, and cross-exchange funding rate arbitrage.
Core Features¶
The plugin provides the following core capabilities:
* One-click mounting: Connects the AlgoVault MCP server to DeepSeek Harness with a single command.
* Composite trading signals: Provides BUY / SELL / HOLD composite trading signals, including confidence and market regime.
* Market regime detection: Identifies market states such as TRENDING_UP, TRENDING_DOWN, RANGING, and VOLATILE.
* Cross-exchange funding rate arbitrage: Scans and returns top-ranked funding rate arbitrage opportunities.
* Track record: Provides real-time performance tracking records.
* Tool namespace: All tools are uniformly prefixed with mcp__algovault__.
Installation and Activation¶
Installing the plugin requires no build step, but pnpm must be available in the system path.
- Add the plugin:
dsh plugin --profile <name> add github:AlgoVaultLabs/dsh-algovault
- Restart the profile:
After installation, the corresponding profile must be restarted for the change to take effect. Plugin membership is read at startup, and hot reloading is not supported.
Typical Usage¶
After the plugin is mounted, tools are exposed in the form mcp__algovault__<tool>. Common tools include:
* get_trade_call: Retrieves the composite trading signal for a single perpetual futures contract.
* scan_trade_calls: Scans top-ranked perpetual futures signals.
* get_market_regime: Retrieves the current market regime and strategy hints.
* scan_funding_arb: Retrieves cross-exchange funding rate arbitrage rankings.
The plugin package includes a skill file (skills/algovault-verdicts/SKILL.md) that guides the model on how to correctly interpret signals returned by the tools. Because this file is not in Harness’s auto-scan path, it must be manually copied to the ~/.dsh/skills/ directory to be loaded.
Configuration and Caveats¶
The free tier can be used without configuration. To unlock higher quotas or the full funding rate result set, configure an API key.
In the Harness profile configuration file (usually cordis.patch.yml), manually add the following configuration. Note: The config field is replaced entirely, so all fields must be redeclared:
- id: mcp-algovault
name: '@deepseek-ai/dsh-mcp-client'
config:
serverName: algovault
transport: streamable-http
url: https://api.algovault.com/mcp?src=dsh-bundle
headers:
Authorization: !!js `Bearer ${process.env.ALGOVAULT_API_KEY}`
Set ALGOVAULT_API_KEY in the system environment variables.
Important limitations:
* Developer preview: DeepSeek Harness is in developer preview. Plugin versions may not be fully aligned with the main release, and breaking compatibility changes may occur.
* HOLD signal: A HOLD signal returned by a tool is a valid trading decision, not a failure. The model should treat it as a stop signal rather than continue looking for other signals.
* Model behavior: The model should not treat signals returned by tools as execution orders. AlgoVault only provides strategy recommendations and does not place orders.
* Performance data: The model should reference real-time data returned by the get_track_record endpoint rather than data from memory.
* Resource bridging: MCP resources and prompts are not bridged by Harness; only tools are provided.
* Skill loading: Built-in skills must be manually copied to ~/.dsh/skills/; automatic discovery is not supported.
Conclusion¶
dsh-algovault provides DeepSeek Harness users with a quick way to access AlgoVault’s trading logic. Developers can build agents using the market data and signal analysis capabilities it offers without modifying code. For more technical details, refer to the project source code or the official integration guide.