@darkiceinteractive/mcp-conductor 1.0.0

package/templates/skill-mcp-conductor.md

@@ -0,0 +1,64 @@
+# MCP Conductor - Token-Efficient Operations
+
+When performing MCP operations, **use mcp-conductor's `execute_code` tool** instead of making direct MCP calls.
+
+## When to Use execute_code
+
+**ALWAYS use execute_code when:**
+- You need 2+ MCP tool calls
+- Processing, filtering, or transforming data from MCP results
+- Iterating over files, repos, search results, or lists
+- Working with large datasets
+
+**Direct calls are OK only for:**
+- Single, simple tool call with no processing
+- Debugging or interactive exploration
+
+## Why This Matters
+
+Direct MCP calls put ALL data in Claude's context:
+- `filesystem.list_directory` → full listing in context
+- `filesystem.read_file` x 10 → 10 full files in context
+- Each response consumes tokens
+
+With execute_code:
+- Write code once (small cost)
+- All intermediate data stays in sandbox (zero cost)
+- Only final return value in context (minimal tokens)
+
+**Savings: 90%+ for multi-step operations**
+
+## Quick API Reference
+
+```typescript
+// Get server client
+const fs = mcp.server('filesystem');
+const github = mcp.github;  // attribute style
+
+// Call tools
+const files = await fs.call('list_directory', { path: '/src' });
+const repos = await github.call('search_repositories', { query: 'ai' });
+
+// Search all tools
+const tools = await mcp.searchTools('file');
+
+// Return only what's needed
+return { count: files.entries.length };
+```
+
+## Example: Process Multiple Files
+
+```typescript
+const fs = mcp.server('filesystem');
+const files = await fs.call('list_directory', { path: '/src' });
+const tsFiles = files.entries.filter(f => f.name.endsWith('.ts'));
+const contents = await Promise.all(
+  tsFiles.map(f => fs.call('read_file', { path: '/src/' + f.name }))
+);
+return {
+  count: tsFiles.length,
+  totalLines: contents.reduce((sum, c) => sum + c.split('\n').length, 0)
+};
+```
+
+This processes potentially hundreds of files but only returns a small summary to context.