Finding the Right MCP Tool: Decision Trees and Execution Strategies for Agents

When operating against a rich Model Context Protocol (MCP) server like Wikantik (which exposes 21+ tools across search, page editing, knowledge graph traversal, and verification), choosing the most cost-effective and context-efficient tool is critical for high agent performance.

This decision tree guide details when to use each MCP tool, how to minimize token burn, and how to execute the 4-tier context escalation ladder.


1. Quick-Reference: MCP Tool Selection Decision Matrix

+-----------------------------------------------------------------------------------------+
|                               MCP TOOL SELECTION MATRIX                                 |
+-----------------------------------------------------------------------------------------+
| Agent Intent           | Recommended Primary Tool          | Fallback / Next Escalation |
+------------------------+-----------------------------------+----------------------------+
| Session Briefing       | `get_briefing`                    | `assemble_bundle`          |
| Broad Topic Research   | `search_pages` (Hybrid search)    | `retrieve_context` (RAG)   |
| Read Exact Article     | `read_pages` (Multi-page batch)   | `get_page`                 |
| Explore Graph Relations| `traverse` / `query_nodes`        | `sparql_query` (Advanced)  |
| Update Existing Page   | `update_page` (With ContentHash)  | `write_pages`              |
| Verify Batch Integrity | `verify_pages` (Batch audit)      | Manual link grep           |
+-----------------------------------------------------------------------------------------+

2. The 4-Tier Context Escalation Ladder

Agent Escalation Ladder:
Tier 1: High-level briefing & bundle -> `get_briefing` / `assemble_bundle`
                | (Insufficient detail)
                v
Tier 2: Targeted semantic search -> `retrieve_context` / `search_pages`
                | (Need full text of cited pages)
                v
Tier 3: Exact page reads in batch -> `read_pages` (up to 5 slugs)
                | (Need complex ontology path)
                v
Tier 4: Jena SPARQL Graph queries -> `sparql_query`