MCP Server
The Oxagen MCP server exposes your workspace knowledge graph as typed, workspace-scoped tools to any MCP-compatible coding agent.
The Oxagen MCP server runs at https://mcp.oxagen.ai and exposes the
canonical Oxagen tool catalog to any MCP-compatible client. For the
full catalog — every tool, with surfaces, credit cost, and policy —
see Agent Tools. For the underlying model — one
registry, three call sites — see Agent Overview.
Every MCP call is workspace-scoped, credit-metered, and audited. The server operates against the same Neo4j-backed graph as the in-app chat and the REST API; an edit made from Claude Code shows up in the dashboard on the next turn.
Installation
Claude Code
claude mcp add oxagen --transport sse https://mcp.oxagen.ai/sseYou will be prompted for your API key on first connection.
Cursor
Add to .cursor/mcp.json in your project root:
{
"mcpServers": {
"oxagen": {
"url": "https://mcp.oxagen.ai/sse",
"headers": {
"Authorization": "Bearer <your-api-key>"
}
}
}
}VS Code (Copilot)
Add to .vscode/mcp.json:
{
"servers": {
"oxagen": {
"type": "sse",
"url": "https://mcp.oxagen.ai/sse",
"headers": {
"Authorization": "Bearer <your-api-key>"
}
}
}
}Windsurf
Add to your Windsurf MCP configuration:
{
"oxagen": {
"serverUrl": "https://mcp.oxagen.ai/sse",
"headers": {
"Authorization": "Bearer <your-api-key>"
}
}
}Codex
Add to your Codex MCP configuration:
{
"mcpServers": {
"oxagen": {
"transport": "sse",
"url": "https://mcp.oxagen.ai/sse",
"headers": {
"Authorization": "Bearer <your-api-key>"
}
}
}
}ChatGPT
ChatGPT has no command-line installer. Connect via the connector settings UI.
- In the Oxagen web app, open Settings → MCP, pick ChatGPT in the install selector, and generate a token. The page surfaces the server URL, the authentication scheme, and the bearer token.
- In ChatGPT, open Settings → Connectors → New Connector.
- Paste the values:
| Field | Value |
|---|---|
| MCP server URL | https://mcp.oxagen.ai/sse |
| Authentication | Bearer |
| Bearer token | oxa_live_… (from the Oxagen UI) |
The token is workspace-scoped — the connector reads and writes the ontology for the workspace the token was minted under, no other.
Authentication
Every tool call requires a bearer token in the Authorization header.
Obtain your API key from the Oxagen web app under Settings → API
Keys.
API keys are workspace-scoped. A key issued for workspace A cannot read or write workspace B.
What's exposed
The MCP server publishes every tool with mcp in its surfaces — see
Agent Tools for the canonical list. As a quick
orientation, the tools the agent reaches for most from an IDE:
- Ontology — read for search, schema introspection, and read-only Cypher against your graph.
- Code graph for symbol lookup, callers / callees, dependency paths, blame, and PR history.
- Agent sessions for swarm coordination — claim a node, record findings, hand off work.
The MCP surface skips two families that only make sense in the dashboard:
- Visualization tools (
viz.*) are app-only — your IDE renders its own UI for graph and table output, so the server doesn't ship interactive renderers. - Confirmation cards are surfaced as
oxagen_confirmcontent blocks; how the client renders the approve/decline prompt is up to the IDE.
Confirmations
Tools whose policy is confirm or confirm_maybe (destructive
ontology writes, code.apply_patch, schedule create/delete) return a
structured oxagen_confirm content block instead of executing
immediately. The block carries a tool_use_id plus a human-readable
summary of the pending action; the client surfaces an approve /
decline UI and resubmits the confirmation via the standard MCP tool
result. See the policies table in
Agent Tools for which tools fall
into which policy.
Credits
Each tool's credit cost is listed in Agent Tools. Credits are reserved before a tool runs and refunded automatically when the handler errors — a failed call costs nothing. Buy credits from the web app at Settings → Billing.
Rate limits
The MCP server enforces per-workspace rate limits:
- Free tier: 60 tool calls / minute
- Pro: 300 tool calls / minute
- Enterprise: configurable
Exceeding the limit returns a 429 Too Many Requests with a
Retry-After header.
Get started free · Read the agent tools catalog · Read the API docs
Artifact storage
Where Oxagen stores everything the agent produces — the unified app.documents table, the source / provider / generated_by_run_id discriminators, and the documents browser that answers "what did the agent ship last week?".
Plugins
Extend your Oxagen workspace with plugins — pre-built integrations that add tools, slash commands, and agent capabilities to your MCP-connected coding agents.