Setup · Open source · MIT

One command per plugin.
Install only what you need.

hypermnesia-mcp (Cortex), session-optimizer, ai-architect-mcp-codebase and ai-architect-mcp-spec each ship from their own repository and Claude Code marketplace. One command adds the marketplace, one command installs the plugin. Codex, Gemini CLI and other MCP hosts have their own install paths in each README.

These commands are for Claude Code. Each plugin installer copies its agents, skills, hooks, and tools into ~/.claude/.

Pick what you need.

Each plugin is useful alone. Installed together, they interoperate through the same MCP layer. Tier 01 is mature; tier 02 is still in development.

Tier 01 · Mature and usable

Stable releases, marketplace or package installs, a test suite that passes today.

Memory · v4.21.0

hypermnesia-mcp (Cortex)

Persistent memory across sessions, distributed as hypermnesia-mcp. 52 MCP tools, 9 lifecycle hooks, 36 cited neuroscience mechanisms. LongMemEval-S Recall@10 97.8% / MRR 0.9046 (measured on v4.20.0). SQLite by default, PostgreSQL + pgvector optional.

> /plugin marketplace add cdeust/Cortex
> /plugin install hypermnesia-mcp

Restart Claude Code, then run /cortex-setup-project to bootstrap the database, embedding model, and hooks. The same marketplace also carries the read-only hypermnesia-mcp-viz (Cortex Viz): /plugin install hypermnesia-mcp-viz@cortex-plugins. Read more →

Visualization · v3.2.0

hypermnesia-mcp-viz (Cortex Viz)

Read-only companion to Cortex, formerly cortex-viz. Six reading angles: Graph, an anatomical 3D brain view, Trace, Knowledge, Wiki, Board. Reads PostgreSQL read-only (Trace also works without it); never writes.

> /plugin marketplace add cdeust/Cortex
> /plugin install hypermnesia-mcp-viz@cortex-plugins

Ships from the same marketplace as Cortex — no separate marketplace add. Read more →

Context budget · v2.3.0

session-optimizer

Three plugins for long Claude Code sessions: context-guard (a Stop hook that enforces a per-model context budget and checkpoints at the warning threshold), refine-gate (portable prompt refinement, also for Codex and Gemini CLI) and statusline (a multi-line status bar tied to the same thresholds).

> /plugin marketplace add cdeust/session-optimizer
> /plugin install context-guard@session-optimizer-marketplace

Same marketplace: refine-gate@session-optimizer-marketplace and statusline@session-optimizer-marketplace. cdeust/session-optimizer

Tier 02 · In development

Public and working, still being improved: pre-1.0 versions. Install them knowing that.

in development Code intelligence · v0.11.1

ai-architect-mcp-codebase

Cross-platform Rust MCP server. Indexes codebases in 11 languages into a property graph, resolves call chains, detects communities, traces execution flows. 26 MCP tools, 1,200+ tests.

> /plugin marketplace add cdeust/ai-architect-mcp-codebase
> /plugin install ai-architect-mcp-codebase@ai-architect-mcp-codebase-marketplace

Requires Rust 1.95.0 (pinned by rust-toolchain.toml; rustup installs it) and CMake on PATH; the release binary builds on first launch (~5 minutes, cached after). Read more →

in development PRD pipeline · v0.8.0

ai-architect-mcp-spec

Multi-judge PRD generation and verification with closed-loop, externally-grounded calibration. 17 MCP tools, 11 PRD steps plus 9 opt-in implementation steps, 1,506 tests. 10 packages, TypeScript/Node.

> /plugin marketplace add cdeust/ai-architect-mcp-spec
> /plugin install ai-architect-mcp-spec@ai-architect-mcp-spec-marketplace

Requires Node.js 20.x or 22.x. The bundled MCP server is self-contained; better-sqlite3 is an optional native dependency that degrades gracefully to in-memory mode when absent. Read more →

Generate your first PRD.

Once ai-architect-mcp-spec is installed, describe a feature in plain language.

1. Start the pipeline

> /generate-prd build a user authentication system with OAuth2, MFA, and session management

The reducer classifies intent, runs codebase discovery (deeper with ai-architect-mcp-codebase installed), and asks clarification questions until you confirm the scope is right.

2. The pipeline runs 11 steps

Banner → preflight → context detection → input analysis → feasibility gate → clarification → budget → section generation (Cortex recall → draft → validate → retry up to 3x) → JIRA ticket synthesis → file export → multi-judge self-check.

3. Nine files are written

FileContents
01-prd.mdProduct requirements document
02-data-model.mdData model
03-api-spec.mdAPI specification
04-security.mdSecurity & performance requirements
05-testing.mdTesting strategy & acceptance criteria
06-deployment.mdDeployment plan, timeline, risks
07-jira-tickets.mdSynthesized JIRA tickets
08-source-code.mdSource code
09-test-code.mdTest code

Plus a 00-run-notes.md naming any skipped section and why. Every numeric claim in the output traces to a citation, a benchmark, or an explicit source: provisional heuristic admission.

What gets installed.

Verified counts per plugin, from each release tree.

PluginVersionShips
hypermnesia-mcp (Cortex)4.21.052 memory MCP tools (55 with the codebase and spec integrations), 9 lifecycle hooks, 36 cited neuroscience mechanisms, a 97-paper bibliography, 6,788 tests passing
ai-architect-mcp-codebase (in development)0.11.126 MCP tools, 11 languages, 1,200+ tests
ai-architect-mcp-spec (in development)0.8.017 MCP tools, 11 PRD steps + 9 opt-in implementation steps, 1,506 tests, 9 output files per run
hypermnesia-mcp-viz (Cortex Viz)3.2.0Formerly cortex-viz: six read-only reading angles (Graph, 3D brain, Trace, Knowledge, Wiki, Board)
session-optimizer2.3.0context-guard 2.0.1, refine-gate 2.1.0, statusline 2.2.0

All are MIT-licensed and free. Figures are sourced to each plugin's own release tree (README badges and figures, .claude-plugin/plugin.json and marketplace version fields), re-verified 14 September 2026.

Requirements.

Two of these plugins need a toolchain on PATH. Cortex needs nothing beyond Claude Code.

PluginPrerequisite
CortexNone — runs on the built-in SQLite backend by default. PostgreSQL 15+ with pgvector is opt-in for larger corpora.
ai-architect-mcp-codebaseRust 1.95.0 (pinned; rustup installs it) and CMake — the release binary builds on first launch, cached after.
ai-architect-mcp-specNode.js 20.x or 22.x. better-sqlite3 is an optional native dependency; the pipeline falls back to in-memory mode without it.

Troubleshooting.

Plugin not found

  • Re-run the marketplace add for that plugin, then /plugin install <name>
  • Restart Claude Code after any install
  • Update Claude Code: claude update

ai-architect-mcp-codebase binary missing

  • Confirm rustc --version reports 1.95.0 inside the plugin directory
  • Build manually: cargo build --release inside the plugin's install path
  • CMake must be on PATH (LadybugDB's C++ core builds from source on first run)

ai-architect-mcp-spec MCP tools not registering

  • Confirm node --version reports 20.x or 22.x
  • Restart Claude Code — the 17 tools register on first stdio handshake
  • better-sqlite3 failing to build is non-fatal: the evidence repository degrades to in-memory mode

Cortex PostgreSQL connection

  • Verify with python3 -m mcp_server.doctor
  • On the default SQLite backend, the PostgreSQL checks report “not set” and can be ignored
  • Re-run /cortex-setup-project to retry pgvector installation and database creation

Still stuck? Contact us or open an issue on the relevant plugin's GitHub repository.

Free & open source

Every plugin here is MIT licensed. Install one, or install them all — each works standalone, all interoperate.

Part of the engagement journey · Exhibit 02

Where this setup fits in an engagement

This is the activation itself — workspace setup, permissions, MCP wiring, and team onboarding for the plugins above, done so your team can run them without us in the room.

Stage Exhibit 02 · Claude Code activation