Oxagen Docs

Extend your Oxagen workspace with plugins — pre-built integrations that add tools, slash commands, and agent capabilities to your MCP-connected coding agents.

Plugins extend the Oxagen MCP server with additional tools, slash commands, and workspace capabilities. A plugin is a versioned manifest that declares what tools it adds, what permissions it requires, and how to install it. Once installed, a plugin's tools appear alongside the core Oxagen tools in any connected MCP client.

Plugins are workspace-scoped. Installing a plugin for one workspace does not affect others.

How plugins work

A plugin manifest declares a set of MCP tools (or slash commands) and the permissions they need (e.g. ontology:read, ontology:write, memory:read).
A workspace owner installs the plugin via the Oxagen web app or API.
The plugin's tools are registered with the workspace's MCP server session.
Any MCP-connected agent (Claude Code, Cursor, VS Code, Windsurf) can call the plugin's tools with the same syntax as core tools.

Plugins run in an ephemeral sandbox (services/worker-sandbox) that is isolated from the main API and worker. Plugins cannot read raw credentials, access other workspaces, or bypass RLS.

Installing a plugin

Via API

POST /v1/plugins accepts the raw YAML manifest body (manifest) plus an optional install-time config map. The plugin is workspace-scoped to the caller's default workspace.

curl -X POST https://api.oxagen.ai/v1/plugins \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{
        "manifest": "name: my-plugin\nversion: 0.1.0\nkind: mcp_tool\n...",
        "config": {}
      }'

Claude Code plugin marketplace

The Claude Code plugin marketplace install flow is the recommended path for human-installable plugins — see Claude Code plugin.

Plugin permissions

Every plugin declares the minimum permissions it needs. You approve them at install time. Permissions are:

Permission	What it allows
`ontology:read`	Read nodes, edges, and types
`ontology:write`	Create and update nodes and edges
`memory:read`	Read agent memory nodes
`memory:write`	Write agent memory annotations
`connections:read`	Read connection metadata (not credentials)
`agent:coordinate`	Start sessions, claim work, record findings

Plugins cannot be granted permissions beyond what the installing user holds in the workspace.

Available plugins

Claude Code plugin

The official Oxagen plugin for Claude Code. Open source at oxagenai/claude-code-plugin (MIT). Ships 50+ slash commands across discovery, traversal, history, tests, CI, security, memory, incident response, and multi-agent coordination. The SessionStart hook pins three guarantees on every turn — graph first, memory always recalled, coordination respected — so the agent treats Oxagen as a first-class concern across coding, incident response, security triage, customer-impact analysis, and research workflows.

View Claude Code plugin docs

Install:

/plugin marketplace add oxagenai/claude-code-plugin
/plugin install oxagen@oxagen
/oxagen:setup

Source: github.com/oxagenai/claude-code-plugin

Building a plugin

Plugins are JSON manifests paired with a serverless function or MCP endpoint. The manifest schema is documented in the API reference. Plugin development tooling (plugin-dev:create-plugin) is available in the Oxagen Claude Code plugin.

Get started free · View on GitHub

Plugins

On this page