Guidance for when to use MCP vs direct API calls
Decision framework for choosing between MCP tools and direct API calls.
Need to call an external service?
โโ Is there a skill for it? โ Use the skill
โโ Is cost control important? โ Use direct API
โโ Need complex multi-step workflow? โ Consider MCP
โโ Simple one-off call? โ Direct API is fine
Prefer skills/direct API when:
| Scenario | Why |
|---|---|
| Cost-sensitive services | Control model selection (Perplexity) |
| Simple queries | Less overhead than MCP |
| High-frequency calls | Faster, no tool discovery |
| Need predictable behavior | Skills have explicit instructions |
Services with skills:
| Service | Skill | Why not MCP |
|---|---|---|
| Perplexity | research |
MCP defaults to expensive models |
| PostHog | analytics |
Direct API is simpler |
email |
Himalaya CLI, no MCP needed | |
| GitHub | github |
gh CLI is more reliable |
Use MCP when:
| Scenario | Why |
|---|---|
| Complex multi-step workflows | MCP handles tool chaining |
| Need tool discovery | Don't know exact endpoint |
| Official MCP server exists | Better maintained |
| Service has many endpoints | MCP abstracts complexity |
Good MCP use cases:
| Service | MCP Server | Use Case |
|---|---|---|
| Apify | apify |
Web scraping (many actors) |
| DataForSEO | dfs-mcp |
SEO data (many endpoints) |
| Prospect | prospect |
Lead enrichment |
| Google Workspace | google-mcp |
Calendar, Sheets, Drive |
Some MCP servers have expensive defaults:
| MCP | Issue | Solution |
|---|---|---|
| Perplexity MCP | Defaults to sonar-deep-research |
Use research skill instead |
MCP tool calls include:
For simple calls, this overhead can exceed the actual API call.
MCP servers (especially npx-based) have cold start delays:
| Factor | Use Skill | Use MCP |
|---|---|---|
| One endpoint | โ | |
| Many endpoints | โ | |
| Cost control needed | โ | |
| Complex workflow | โ | |
| Speed critical | โ | |
| Tool discovery needed | โ | |
| Explicit instructions exist | โ |
Check skills/ directory or use /list-skills for current skills.
| Skill | Purpose |
|---|---|
| research | Perplexity web search |
| analytics | PostHog queries |
| Email via Himalaya | |
| github | GitHub via gh CLI |
| lead-handler | Process incoming leads |
Check workspace/mcporter.json for configured MCP servers.
Current servers:
apify - Web scraping actorsdfs-mcp - DataForSEOprospect - Lead enrichmentgoogle-mcp - Google Workspaceโ mcp__perplexity__search("latest AI news")
โ May use sonar-deep-research ($0.40+)
โ Use research skill with sonar model
โ Uses sonar ($0.006)
โ mcp__apify__call-actor for complex scraping
โ Many actors, MCP handles discovery