Source
Agent UX report (2026-02-10) + discussion about useful conceptual framing.
Problem
The MCP tool descriptions lack conceptual framing that helps agents understand how to use the platform effectively. Three terms emerged from a live agent session that would make the tooling more intuitive:
Terms to incorporate
Exocortex
The knowledge graph isn't a passive database — it's an external cognitive system that agents read from and write back to, extending their reasoning across sessions. This framing explains why graph edit, ingestion, and program tools exist alongside search: agents are building persistent reasoning structures, not just querying data.
System 1 / System 2
The platform has two cognitive modes:
- System 1 (reflexive):
search, connect, related — fast associative exploration, "what's nearby?"
- System 2 (deliberate):
program with Cypher — structured multi-step reasoning, "show me the argument structure"
Agents need to develop the instinct for when to switch. The current descriptions don't make this distinction, leading agents to repeat System 1 calls hoping for System 2 results.
Desire paths
connect follows desire paths — the most-trafficked routes through the graph. In a young graph or sparse region, these are structural paths (Source→Ontology→Source). Manually created edges are like freshly laid sidewalks nobody's walked on yet. This single metaphor explains the #1 confusion point from the UX report.
Where to apply
- Server guide comment (L97-128 in mcp-server.ts) — reframe the three tiers using System 1/2 language, mention exocortex
search tool description — "Your entry point to the exocortex" + System 1 framingconcept connect action — desire paths explanation, when to switch to programprogram tool description — "System 2 reasoning tool" framing, emphasize this is for deliberate multi-step thinkingexplore-graph prompt — teach the mode-switching instinct
Related
Source
Agent UX report (2026-02-10) + discussion about useful conceptual framing.
Problem
The MCP tool descriptions lack conceptual framing that helps agents understand how to use the platform effectively. Three terms emerged from a live agent session that would make the tooling more intuitive:
Terms to incorporate
Exocortex
The knowledge graph isn't a passive database — it's an external cognitive system that agents read from and write back to, extending their reasoning across sessions. This framing explains why graph edit, ingestion, and program tools exist alongside search: agents are building persistent reasoning structures, not just querying data.
System 1 / System 2
The platform has two cognitive modes:
search,connect,related— fast associative exploration, "what's nearby?"programwith Cypher — structured multi-step reasoning, "show me the argument structure"Agents need to develop the instinct for when to switch. The current descriptions don't make this distinction, leading agents to repeat System 1 calls hoping for System 2 results.
Desire paths
connectfollows desire paths — the most-trafficked routes through the graph. In a young graph or sparse region, these are structural paths (Source→Ontology→Source). Manually created edges are like freshly laid sidewalks nobody's walked on yet. This single metaphor explains the #1 confusion point from the UX report.Where to apply
searchtool description — "Your entry point to the exocortex" + System 1 framingconceptconnect action — desire paths explanation, when to switch to programprogramtool description — "System 2 reasoning tool" framing, emphasize this is for deliberate multi-step thinkingexplore-graphprompt — teach the mode-switching instinctRelated