AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
MCP verified Apache-2.0 Self-run

Nexusos

mcp-asimons81-nexusos · by asimons81

Files you own. Memory your agents can trust. Also: your AI won't gaslight you about what's in your own notes anymore.

No reviews yet
0 installs
0 views
view→install

Install

$ agentstack add mcp-asimons81-nexusos

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/mcp-asimons81-nexusos)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
yesterday

Declared compatibility

Claude CodeClaude DesktopCursorWindsurf

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Nexusos? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Files you own. Memory your agents can trust.

NexusOS

NexusOS is a local-first knowledge operating system for AI agents. It turns ordinary folders of Markdown and text files into a structured, searchable memory layer exposed through a CLI and Model Context Protocol server.

Your source files remain the system of record. NexusOS builds disposable derived state inside .nexusos/, then gives humans and agents deterministic tools to search, browse, read, inspect links, assemble context, and verify workspace health.

> [!IMPORTANT] > NexusOS v0.1.0 is the first stable release. The planned v0.1 core feature > scope, release hardening, packaging validation, cross-platform proof, > security review, and public contract freeze are complete. See the > [release roadmap](ROADMAP.md).

Why NexusOS

Agent memory should not require surrendering your notes to a proprietary database or trusting an opaque retrieval pipeline.

NexusOS is built around a smaller contract:

  • Local first: core workflows require no hosted account or network connection.
  • Files stay yours: Markdown and text remain readable without NexusOS.
  • Deterministic retrieval: SQLite FTS5, stable IDs, source paths, and line ranges make

results inspectable.

  • Agent native: the same service layer powers the CLI and MCP tools.
  • Read-only by default: retrieval never edits source documents.
  • Rebuildable state: the index can be deleted and regenerated from source files.

Where NexusOS fits

NexusOS is not designed as a self-evolving agent-memory system. It is a source-grounded knowledge substrate: your files remain canonical while NexusOS builds deterministic, rebuildable retrieval state around them.

Read the technical research behind this mapping →

> This is qualitative architectural positioning, not a benchmark or ranking.

The key distinction is authority: NexusOS protects and exposes canonical knowledge, while evolving memory systems may create, update, consolidate, or rewrite memory representations over time.

What works today

v0.1.0 includes:

  • safe workspace initialization with blank and starter templates
  • path boundaries, deny paths, nested-workspace protection, and doctor checks
  • deterministic Markdown and plain-text indexing into SQLite with FTS5
  • incremental indexing and content-aware stale-index detection
  • ranked search with source paths, headings, snippets, and line ranges
  • browse, read, recent, links, and deterministic context navigation
  • workspace linting for link, frontmatter, structure, and staleness problems
  • MCP over stdio and loopback-first Streamable HTTP
  • a read-only local inspection API and bundled UI
  • JSON output for automation-friendly command paths

Not included in v0.1: embeddings, vector search, ingestion connectors, guarded source writes, cloud hosting, OAuth, sync, or multi-user collaboration.

Quick start

Install the stable release from PyPI (full instructions: [docs/install.md](docs/install.md)):

pip install nexusos

nexusos version
nexusos init ./example-workspace
nexusos doctor --workspace ./example-workspace
nexusos index --workspace ./example-workspace
nexusos status --workspace ./example-workspace
nexusos browse --workspace ./example-workspace
nexusos search "workspace" --workspace ./example-workspace

For a disposable end-to-end walkthrough:

nexusos demo

For development from source:

git clone https://github.com/asimons81/nexusos.git
cd nexusos
uv sync
uv run nexusos version

How it works

Markdown and text files
          │
          ▼
  discovery + parsing
          │
          ▼
 deterministic SQLite index
          │
          ├── CLI search and navigation
          ├── workspace linting
          ├── local inspection API and UI
          └── MCP tools for agents

The index is derived state. Source documents are never converted into a proprietary format and can be recovered without NexusOS because they never left the filesystem.

CLI

| Command | Purpose | |---|---| | nexusos version | Print the installed version | | nexusos init PATH | Create or adopt a workspace | | nexusos doctor | Validate workspace health and configuration | | nexusos config show | Display raw or effective configuration | | nexusos index | Build or incrementally update the index | | nexusos status | Report index state, counts, and staleness | | nexusos search TERM | Run ranked FTS5 search | | nexusos browse | List indexed documents | | nexusos read ITEM | Read a document by ID, path, or name | | nexusos recent | List recently modified documents | | nexusos links ITEM | Inspect incoming and outgoing wiki-link state | | nexusos context ITEM | Build a deterministic evidence packet | | nexusos lint --workspace PATH | Lint a workspace vault | | nexusos mcp | Start the MCP server over stdio | | nexusos serve --transport streamable-http | Start MCP over HTTP | | nexusos serve --workspace PATH | Start the inspection API and UI | | nexusos demo | Run a synthetic end-to-end walkthrough |

Use nexusos COMMAND --help for the current option contract. Workspace commands detect the current workspace unless --workspace is supplied.

The CLI commands, options, exit codes, configuration keys/environment variables, JSON shapes, and MCP tool schemas are frozen for the v0.1 train and inventoried in [docs/contracts.md](docs/contracts.md). tests/contracts/ locks the surface; changes require a deliberate roadmap decision and changelog entry.

MCP for agents

Start NexusOS as a stdio MCP server:

nexusos mcp --workspace /path/to/workspace

Generic client configuration:

{
  "mcpServers": {
    "nexusos": {
      "command": "nexusos",
      "args": ["mcp", "--workspace", "/path/to/workspace"]
    }
  }
}

Available tools:

| Tool | Contract | |---|---| | status | Index status, counts, and staleness reasons | | search | Ranked full-text search | | browse | Indexed document metadata | | read | Bounded source reading by ID, path, or name | | recent | Recently modified documents | | links | Incoming and outgoing wiki-link state | | context | Deterministic headings, siblings, and linked evidence | | index | Refresh derived state inside .nexusos/ |

All retrieval tools are read-only. index writes only disposable derived state.

MCP Streamable HTTP is loopback-first but unauthenticated. Do not expose it directly to an untrusted network. See [docs/mcp.md](docs/mcp.md) and [SECURITY.md](SECURITY.md).

Workspace layout

A starter workspace uses a practical folder convention, not a proprietary storage format:

workspace/
├── nexusos.toml
├── README.md
├── SCHEMA.md
├── inbox/
├── raw/
│   ├── articles/
│   ├── conversations/
│   ├── notes/
│   └── transcripts/
├── wiki/
│   ├── concepts/
│   ├── entities/
│   ├── projects/
│   ├── queries/
│   └── _archive/
├── ops/
│   ├── decisions/
│   ├── sops/
│   └── workflows/
├── mocs/
├── journal/
└── .nexusos/          # generated state, safe to rebuild

Collections and file patterns are configurable in nexusos.toml.

Safety boundary

NexusOS v0.1 is designed for a local, single-user workspace controlled by the operator.

The current contract includes:

  • no source-document mutation during indexing, retrieval, linting, or MCP reads
  • denied-path and workspace-boundary validation
  • nested-workspace prevention
  • transactional index writes and an exclusive writer lock
  • temporary-file hardening for critical state writes
  • loopback defaults for local HTTP transports
  • Host validation, Origin checks, and a per-process token for the inspection API

The inspection API and MCP Streamable HTTP are separate surfaces with different security contracts. The inspection API is token-protected and warns on a non-loopback bind; MCP Streamable HTTP is unauthenticated and refuses a non-loopback bind unless the operator explicitly opts in with --allow-non-loopback / NEXUSOS_ALLOW_NON_LOOPBACK=1. A non-loopback bind is not a claim that NexusOS is an internet-facing authenticated service.

Review [SECURITY.md](SECURITY.md) and the active hardening work in [ROADMAP.md](ROADMAP.md) before using NexusOS outside the supported local boundary.

Configuration

Workspaces are configured through nexusos.toml. Effective values are resolved in this order, with later layers overriding earlier ones:

  1. built-in defaults
  2. nexusos.toml
  3. NEXUSOS_* environment variables
  4. CLI flags where supported
nexusos config show
nexusos config show --effective
nexusos config show --json

See [docs/configuration.md](docs/configuration.md) for valid keys, defaults, environment variable names, and current limitations.

Development

uv sync
uv run ruff check .
uv run ruff format --check .
uv run mypy src
uv run pytest -q --cov=nexusos
uv run nexusos version

Coverage is enforced: the aggregate gate is 80% ([tool.coverage.report] fail_under), and security-critical modules (core/path_safety.py, indexing/lock.py, services/serve_service.py, mcp/) keep targeted floors in CI. The full test suite (unit + integration + security) must pass on every supported platform before release.

Read these before changing the repository:

  • [AGENTS.md](AGENTS.md): non-negotiable constraints and task protocol for coding agents
  • [CONTRIBUTING.md](CONTRIBUTING.md): contributor setup and pull-request expectations
  • [docs/architecture.md](docs/architecture.md): dependency direction and system contracts
  • [ROADMAP.md](ROADMAP.md): versioned release tasks and acceptance gates
  • [docs/releasing.md](docs/releasing.md): release procedure and evidence checklist

Agent execution contract

Roadmap work should reference a task ID such as A3-04 or RC-03. Agents must:

  1. state the task and acceptance criteria they are implementing
  2. inspect implementation and tests before editing behavior or docs
  3. preserve architecture boundaries and source immutability
  4. add or update tests for behavioral changes
  5. run the complete verification gate
  6. update affected docs and changelog entries in the same change
  7. report evidence, limitations, and deferred work explicitly

“Implemented” without verification evidence is not a completed roadmap task.

Documentation

| Document | Contents | |---|---| | [ROADMAP.md](ROADMAP.md) | Executable plan from alpha to stable | | [docs/install.md](docs/install.md) | Supported environments, dependencies, install/upgrade, verified artifacts | | [docs/releases/v0.1.md](docs/releases/v0.1.md) | v0.1 release notes (features, fixes, known issues, verification) | | [docs/architecture.md](docs/architecture.md) | Layers, dependencies, and invariants | | [docs/agent-memory-landscape.md](docs/agent-memory-landscape.md) | Research-backed positioning in the agent memory landscape | | [docs/contracts.md](docs/contracts.md) | Frozen CLI, config, JSON, exit-code, and MCP contracts | | [docs/configuration.md](docs/configuration.md) | TOML schema, environment variables, precedence | | [docs/mcp.md](docs/mcp.md) | MCP tools, transports, and client setup | | [docs/linting.md](docs/linting.md) | Workspace and developer lint modes | | [docs/releasing.md](docs/releasing.md) | Build, validation, and release procedure | | [SECURITY.md](SECURITY.md) | Supported threat boundary and reporting | | [CHANGELOG.md](CHANGELOG.md) | Version history |

Release status

The repository has completed the alpha and release-candidate train and is shipping v0.1.0 as the first stable release. The full-suite platform matrix, measured coverage policy, signed-tag gate, and package verification are in place.

Follow progress in [ROADMAP.md](ROADMAP.md).

License

Apache-2.0. See [LICENSE](LICENSE).

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.