MCP Server
Praxrr exposes a Model Context Protocol (MCP) server so AI assistants and automation tools can inspect your configuration and observability state using natural-language workflows. It is read-only: it can query and preview, but it never writes to your Arr instances or PCD databases.
Endpoint
Section titled “Endpoint”The MCP server is served over Streamable HTTP at:
https://<your-praxrr-host>/api/v1/mcpIt speaks JSON-RPC 2.0 and is stateless — each request is self-contained. The
endpoint is enabled by default; set MCP_ENABLED=0 (or false/no/off) to
disable it.
Authentication
Section titled “Authentication”The MCP endpoint uses Praxrr API tokens. There is no OAuth or
Authorization: Bearer support. Create a token in Settings → Security with
the mcp scope (or admin, which includes it), copy it when it is shown (it is
shown only once), and configure your client to send the header with the full
token value (it already starts with prx_):
X-Api-Key: <your token>Notes:
- Tokens work under
AUTH=onandAUTH=oidc, and for remote clients underAUTH=local. - Under
AUTH=off, or underAUTH=localfrom a local IP, the endpoint is reachable without a token (the deployment already trusts the caller/network). - A token without the
mcpscope gets 403. Aread-only token is not enough. - Revoke the token in Settings → Security to cut off a client immediately.
Connecting a client
Section titled “Connecting a client”Point any Streamable-HTTP MCP client at the endpoint and add the X-Api-Key
header. For clients that only speak stdio (for example some desktop assistants),
bridge with mcp-remote:
npx mcp-remote https://<your-praxrr-host>/api/v1/mcp \ --header "X-Api-Key: <your token>"What is exposed
Section titled “What is exposed”- Tools —
list_instances,get_config_health,get_security_posture,get_drift_status,list_databases,list_resolved_entities,get_resolved_entity,search_sync_history, andpreview_sync(a write-free dry run). - Resources —
praxrr://arr-instances,praxrr://drift/summary,praxrr://config-health,praxrr://security-posture,praxrr://databases, plus templated resources for individual instances and resolved PCD entities. - Prompts — ready-made templates for diagnosing drift, reviewing security posture, planning a sync, and explaining a PCD entity.
Credentials are always redacted: instance API keys are exposed only as a fingerprint, and PCD git tokens are never returned. No write or apply operations are available.