How-to · MCP
How to operate the vem MCP server for agents
Keep agents aligned with tasks, decisions, and context packs through explicit tool calls.
Expose the right tools
The MCP server provides structured tools for tasks, decisions, and search so agents do not rely on ad-hoc parsing.
This makes agent behavior predictable and keeps memory updates consistent across sessions.
Treat MCP as the contract between your agents and your project memory.
- get_active_tasks
- update_task
- search_memory
Treat updates as commits
Encourage agents to record evidence with every update and rely on Git for verification.
That workflow keeps memory trustworthy for both humans and automation.
The most valuable updates connect tasks to the code that completed them.
Operational guardrails
Limit which tools are available in production environments and review tool call logs regularly.
Use clear naming conventions for tasks and decisions so agents can retrieve them without ambiguity.
- Least-privilege tool access
- Tool call audit logs
- Consistent naming patterns