maestro-nerve 0.1.0

Shared intelligence memory plane for AI agents — remote MCP + CLI + Claude Code / Codex plugins.

maestro-nerve

Shared intelligence memory plane for AI agents.

One workspace-scoped, evidence-grounded memory. Reachable from Claude Code, Claude Desktop, Codex, ChatGPT, or your own MCP client.

---

maestro-nerve is a persistent, workspace-scoped memory substrate for AI agents. Instead of letting every agent (Claude Code / Claude Desktop / Codex / ChatGPT / ...) keep its own private memory silo, Nerve gives all of them one shared, evidence-grounded memory plane reachable over MCP.

The seven V1 tools are workspace-bound, read-first, and grounded in source evidence:

Tool	What it does
search_workspace	Ranked semantic search across entities, claims, and evidence.
list_entities	Paginated entity browser.
inspect	Aliases + claims + evidence for one entity.
profile	Narrative briefing synthesized from grounded memory.
ground	Evidence records for one claim.
discover	Time-axis feed of what changed.
act.draft	Draft an action (email, reply, task) from workspace memory — never executes.

Plus server_info — a free health probe.

Install (pick one)

You need a Maestro API key first — mint one at https://nerve.maestro.onl/access.

Option A — pip + remote MCP (recommended)

pip install maestro-nerve
mnerve login                                # paste your API key
mnerve access --client claude-code --format mcp-json
# Copy the printed JSON into Claude Code MCP settings. Done.

Supported clients today: claude-code, codex, chatgpt. Formats: envelope (default, scripting-friendly), mcp-json, settings-json, env.

Option B — Claude Code plugin

/plugin marketplace add maestro-ai-stack/maestro-nerve
/plugin install maestro-nerve@maestro-nerve
/reload-plugins

Plugin brings the skill (agent guidance) and the CLI install hint. You still need pip install maestro-nerve && mnerve login for the CLI.

Option C — Codex plugin

codex plugin add maestro-ai-stack/maestro-nerve

Then pip install maestro-nerve && mnerve login && mnerve access --client codex.

Option D — Skill only (no CLI)

npx skills add maestro-ai-stack/maestro-nerve -y -g

Best for agents that only need the skill guidance and already have an API key in their environment.

CLI surface

mnerve login                              # paste / prompt for an API key
mnerve logout                             # remove local credentials
mnerve whoami                             # backend confirms identity + primary workspace
mnerve workspace list                     # list workspaces reachable with this key
mnerve status                             # one-line reachability probe
mnerve access --client <c> --format <f>   # print MCP config for host c, shape f
mnerve version                            # installed package version

mnerve access --format values:

--format	Shape printed on stdout
envelope (default)	{client, transport, workspace, endpoint, config, note} — good for scripts
mcp-json	Bare {mcpServers: {...}} — paste into Claude Desktop / Claude Code
settings-json	Same as mcp-json; merge under mcpServers in your settings.json
env	KEY=value lines; source into the shell that launches the host

Hints are printed to stderr so stdout stays parseable.

Authentication model

• API key lives at ~/.config/maestro/auth.json (platform-appropriate), mode 0600.
• Override path with MAESTRO_AUTH_FILE=/path/to/auth.json.
• The key is a bearer token for https://nerve-api.maestro.onl/api/*. Workspace isolation is enforced server-side; the key cannot reach workspaces it does not own.
• mnerve logout removes the local file but does not revoke the key server-side. Revoke at https://nerve.maestro.onl/access.

Remote MCP endpoint

POST https://nerve-api.maestro.onl/api/mcp?workspace=<your-slug>
Authorization: Bearer <your api key>
Content-Type: application/json

Every hosted MCP request is workspace-scoped at the transport. You cannot leak between workspaces with one key.

Development / internal backend

The backend (pipeline, extraction, Brain UI) is closed source and lives in a private sibling repo. This public repo is the client-only distribution surface. Issues, feature requests, plugin changes: file here. Backend-specific concerns: file here too — we triage and route internally.

License

MIT — see LICENSE.

Built by Maestro — Singapore AI product studio.