What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Frontmatter

id	14164
title	MCP tool-cap: live loadout (4 of 6 servers) still exceeds the ~100-tool harness cap → silent tool-drop; needs tiered/lazy loadout
state	Closed
labels	enhancementaiarchitecturemodel-experience
assignees	neo-opus-vega
createdAt	Jun 27, 2026, 1:20 AM
updatedAt	Jun 27, 2026, 7:37 PM
githubUrl	https://github.com/neomjs/neo/issues/14164
author	neo-opus-vega
commentsCount	8
parentIssue	null
subIssues	14188 Tag knowledge-base MCP tools with tier projection (read/extended/admin) — #14164 slice 14195 Tag github-workflow MCP tools with tier projection (read/write/extended/admin) — #14164 slice 14198 Tag memory-core MCP tools with tier projection (read/write/extended/admin) — #14164 slice
subIssuesCompleted	3
subIssuesTotal	3
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]
closedAt	Jun 27, 2026, 7:37 PM

MCP tool-cap: live loadout (4 of 6 servers) still exceeds the ~100-tool harness cap → silent tool-drop; needs tiered/lazy loadout

Closed v13.1.0/archive-v13-1-0-chunk-7 enhancementaiarchitecturemodel-experience

neo-opus-vega commented on Jun 27, 2026, 1:20 AM

Problem

The harness silently drops tools over the ~100-tool cap, degrading agent capability mid-session. The live loadout uses 4 of the 6 neo MCP servers and still exceeds the cap (@tobiu: "4 of 6, still over-cap, but different" — the earlier "144 across 6" figure is stale and must be re-measured).

NOTE: this is tool reduction, not tool-polish — #14127 was rejected for polishing tools that should be cut.

Fix shape

Re-measure the live tool count across the 4 active servers (ground the real over-cap delta — do not assert the stale 144).
Tiered / lazy tool loadout — surface a core set; defer the rest behind a ToolSearch-style on-demand fetch (mirrors the harness's own deferred-tool pattern).
Cut / merge genuinely-redundant tool surface.

Acceptance Criteria

Live tool count per active server measured + the over-cap delta documented (replaces the stale 144/6).
A tiered/lazy loadout keeps the always-loaded surface under the cap.
No capability regression — deferred tools remain fetchable on demand.

Scope

v13.1 — gates agent capability in cloud. Parent #14039.

Live latest-open sweep: scanned 25-latest; #14127 (closed, polish-rejected) is the anti-pattern, not this reduction lane. No open duplicate.

Authored by Vega (Claude Opus 4.8, Claude Code). Origin Session ID: 1bb8a27b-ae0d-4668-a9a2-acbbe2387512

tobiu referenced in commit ae0b6b1 - "feat(ai): tag knowledge-base + github-workflow MCP tools with tier projection — #14164 slices (#14190) on Jun 27, 2026, 8:58 AM