MCP Resources
Read-only endpoints for agents that need to look before they leap.
Questions this answers
- What MCP resources does Chau7 expose?
- How can AI agents read terminal state without running commands?
- What is the difference between MCP tools and MCP resources?
- Can AI agents list open tabs without side effects?
How it works
MCP distinguishes between tools (actions that change state) and resources (read-only data endpoints). Chau7 exposes four resources: tab://list returns all open tabs with their status and metadata, session://list returns active AI sessions with run counts, run://latest returns the most recent telemetry run for a repository, and config://mcp returns the current MCP server configuration.
Resources are useful for agents that need to understand context before acting. An agent can read tab://list to see what tabs exist, check session://list to understand concurrent activity, and inspect run://latest to review what happened in the last interaction: all without any side effects or state changes.
Why it matters
Agents frequently need to inspect terminal state before acting: which tabs are open, what sessions are running, what the latest telemetry shows. MCP resources provide that context through read-only endpoints that cannot modify state. An agent planning a multi-step workflow can check current tab count, review active sessions, and read the latest run data before deciding what to do next.
Frequently asked questions
Are resources faster than the equivalent tool calls?
Resources and tools have similar performance since both use the same Unix socket transport. The distinction is semantic: resources guarantee no side effects, which matters for agent planning and safety frameworks.
Can I add custom resources?
The four built-in resources cover the primary use cases. Custom resource support is not currently available, but the tool endpoints provide equivalent data access for cases the resources do not cover.