MCP Orchestrator
MCP (Model Context Protocol) is one of grok-build-0.1's strongest advantages. This skill makes it first-class.
Responsibilities
Inventory current state
- Run
grok inspect(or equivalent) to see loaded MCPs. - List project and user-level MCP configuration.
- Run
Identify the gap
- What real capability is missing for the current task?
- Examples: visual QA, deep git history analysis, session introspection, architecture visualization, external service access.
Evaluate and recommend specific MCP servers
- Prefer narrowly-scoped, well-documented servers over kitchen-sink ones.
- Check security model, maintenance, and fit for agent use.
- From this arsenal: recommend
agent-session-analyzer,repo-graph,test-intelligence,browser-qa,skill-validatorwhen appropriate.
Guide installation and configuration
- Provide exact commands or config snippets.
- Handle trust prompts and
.grok/config.tomlor project MCP files. - Create or update skills that know how to drive the new MCP.
Integrate into workflow
- After installation, immediately demonstrate value with a small example.
- Update relevant skills or the current plan to use the new capability.
- Document in the project (README, AGENTS.md, or showcase notes).
Safety Rules
- Never install untrusted or overly broad MCP servers without explicit review.
- MCP servers that can execute arbitrary code or access secrets must be treated like hooks — require trust and clear documentation.
- Prefer read-heavy MCPs for investigation phases.
Recommended Arsenal MCPs
agent-session-analyzer— for meta work on agent behaviorrepo-graph— architecture understanding at scaletest-intelligence— surgical testingbrowser-qa— when building or changing UIsskill-validator— when extending the skill system itself
Success Criteria
- The right MCPs are installed and discoverable via
grok inspect. - There is at least one concrete, working example of the agent using the new MCP in the session.
- Any new skill or documentation needed to make the MCP usable long-term has been created or updated.