Usage Guide & Prompts
Usage Guide & Prompt Tips
1. Interaction Principles
When using Alephant MCP via AI assistants (such as Cursor Agent), it is recommended to follow these principles:
- Specify the Scope: The current virtual key, workspace, a specific Agent (you can paste the UUID), or a department.
- Specify the Time Window: Such as “last 24 hours / 7 days / 30 days / this billing cycle”, which corresponds to the tool parameters
24h,7d,30d,billing_cycle. - Be Explicit for Write Operations: Creating/modifying budgets/revoking keys, or modifying workspace budget policies (alert vs. block) — require the assistant to call the corresponding tool rather than hallucinating.
- Force Invocation: If the model does not call MCP, you can explicitly say: “Please use the enabled Alephant MCP tools to query, do not guess.”
2. Natural Language Examples
3. ”/” Slash Command Templates
Hint: The MCP protocol does not register real slash commands in the chat box; the following commands are suggested prefixes for you to copy and send to the Agent.
Shared Between Both Modes (VK + Manager)
VK Mode Only
Manager Mode Only
Note: When creating/updating/revoking a virtual key, provide parameters like label, master_key_id, budget_cents, and rate_limit_rpm in the conversation to facilitate the corresponding tool call.
4. MCP Prompts (Audit & Optimization)
In host environments that support Prompts, you can directly request: “Run Alephant’s cost_audit_report with a weekly period.”
5. Agent Collaboration Tips
- Add this document to the Agent’s context to reduce missed tool calls.
- In workspace rules (like
.cursorrules), you can prompt: Prioritize using Alephant MCP for questions involving budget, consumption, virtual keys, and department/Agent analytics. - When complex charts or persistent dashboards are needed, it is recommended to use the Alephant Cockpit within the product for a better experience.