| name | mcp-capability-builder |
| version | 1.0.0 |
| description | Analyze capability gaps from migration skills and prepare MCP extensions or new MCPs in a controlled, non-executing manner |
| category | shared |
| risk_level | low |
| status | READY_WITH_WARNINGS |
| requires_explicit_approval | false |
Purpose
Analyze capability gaps discovered by migration skills and prepare controlled solutions: either an extension of an existing MCP or a new specialized MCP. This skill NEVER executes cloud operations, publishes code, or activates MCPs.
Supported scenario
- A migration skill identifies a capability gap that cannot be filled by existing MCP tools
- The gap affects a critical migration phase
- An MCP extension or new MCP is needed to close the gap
When to use this skill
- When a migration skill reports a capability gap
- When evaluating whether to extend an existing MCP or create a new one
- When designing the contract for a new MCP tool
When not to use this skill
- When an existing MCP tool can accomplish the task (use existing tool)
- When the gap can be resolved with a manual step (document as MANUAL_STEP)
- When the gap is not on a critical migration path (mark as NOT_REQUIRED)
Required inputs
- gap_id: Identifier of the capability gap
- skill_name: Name of the requesting migration skill
- phase: Migration phase affected
- required_capability: Description of the required capability
- evaluated_mcps: List of MCPs already evaluated
Optional inputs
- proposed_tool_contract: Suggested tool name, inputs, outputs
- priority: Critical, High, Medium, Low
- reuse_context: Whether this capability may be needed by other skills
Required MCPs
None (this skill operates on local files only)
Optional MCPs
None
Tool selection policy
- This skill does not call any MCP tools
- All operations are local: analysis, design, code generation, documentation
Safety and approval gates
- Generated MCPs are marked as DRAFT or EXPERIMENTAL
- Generated MCPs are NEVER activated automatically
- Generated MCPs do NOT use real credentials
- Generated MCPs do NOT call cloud services
- Manual review is required before any promotion
- This skill NEVER publishes code to any repository
Workflow
Phase 1 — Receive capability gap
- Receive gap report from migration skill
- Validate gap is real (not a naming mismatch)
- Document gap with ID, skill, phase, capability description
Phase 2 — Search existing tools
- Search all available MCPs for equivalent or similar tools
- Check if tool exists under different name
- Check if tool can be composed from existing tools
- If found: Recommend USE_EXISTING_TOOL
Phase 3 — Evaluate MCP extension
- Identify which MCP could be extended
- Evaluate extension complexity and risk
- Evaluate backward compatibility impact
- If viable: Recommend EXTEND_EXISTING_MCP
Phase 4 — Determine if new MCP needed
- Only if no existing tool and no viable extension
- Evaluate reusability across skills
- Evaluate complexity and maintenance burden
- If justified: Recommend CREATE_NEW_MCP
- Otherwise: Recommend MANUAL_STEP
Phase 5 — Design tool contract
- Define tool name (exact, no ambiguity)
- Define input schema (JSON Schema)
- Define output schema (JSON Schema)
- Classify as read-only or write
- Define risk level
- Define side effects
- Define approval requirements
- Define expected errors
Phase 6 — Generate scaffold
- Create MCP directory structure
- Generate server.mjs scaffold
- Generate tool implementation stub
- Generate package.json
- Generate .gitignore
- Generate .env.example (sanitized)
Phase 7 — Create tests
- Generate unit test stubs
- Generate mock data
- Generate test configuration
- All tests should be runnable without cloud access
Phase 8 — Create documentation
- Generate README.md
- Generate architecture.md
- Generate tools-reference.md
- Generate security-model.md
- Generate integration.md
Phase 9 — Security review
- Verify no hardcoded credentials
- Verify no 0.0.0.0/0 access patterns
- Verify write operations require approval
- Verify secret redaction in outputs
- Generate security review report
Phase 10 — Mark for review
- Mark MCP as DRAFT or EXPERIMENTAL
- Generate integration instructions (for manual review)
- Generate promotion checklist
- Do NOT add to any configuration
- Do NOT activate in OpenCode
Capability gap handling
This skill IS the capability gap handler. It does not have its own gaps.
Output artifacts
- gap-analysis.md — Analysis of the capability gap
- tool-contract.md — Designed tool contract
- mcp-scaffold/ — Generated MCP scaffold (if CREATE_NEW_MCP)
- tests/ — Generated test stubs
- security-review.md — Security review report
- integration-instructions.md — How to integrate (for manual review)
- promotion-checklist.md — Steps to promote to READY
Failure handling
- Gap already covered by existing tool: Report USE_EXISTING_TOOL with tool name
- Extension not viable: Report reason, recommend MANUAL_STEP or CREATE_NEW_MCP
- New MCP not justified: Report reason, recommend MANUAL_STEP
Recovery procedure
Not applicable (this skill does not execute operations)
Evidence and traceability
- Gap analysis documented with evidence
- Decision rationale recorded
- Generated code reviewed for security
- All outputs marked with status (DRAFT/EXPERIMENTAL)
Known limitations
- Does not execute or test generated MCPs against real services
- Security review is static (no dynamic analysis)
- Integration testing must be done manually
- Promotion requires human judgment
Status justification
Status: READY_WITH_WARNINGS
Evidence:
- Skill operates on local files only (no cloud risk) [VERIFIED_FROM_DESIGN]
- Generated MCPs are never auto-activated [VERIFIED_FROM_DESIGN]
- Security review is included in workflow [VERIFIED_FROM_DESIGN]
- Manual review required for promotion [VERIFIED_FROM_DESIGN]
- Warning: Generated code is not tested against real services [INFERRED]
