Skip to main content
GET /system-prompt returns the full per-session system prompt as plain text. It’s Tier 6 of the six system-prompt tiers — kept because the Open WebUI filter needs it, and because it’s the simplest integration for anything that isn’t an MCP client.

​
Request

curl "http://localhost:8081/system-prompt?chat_id=my-session&user_email=user@example.com"

​
Priority

The server resolves chat_id and user_email from headers first, then query params, then defaults: 
X-Chat-Id | X-OpenWebUI-Chat-Id     > ?chat_id=           > "default"
X-User-Email | X-OpenWebUI-User-Email > ?user_email=     > None
This matches how the rest of the server reads session identity.

​
Response

  • Body — rendered system prompt text. Includes <available_skills>, file URL hints, skill mount paths, and the per-chat identity block.
  • Header — X-Public-Base-URL: <PUBLIC_BASE_URL>. The Open WebUI filter caches this alongside the prompt so its outlet() can build browser-facing archive/preview links without its own Valve.

​
Cache

render_system_prompt(chat_id, user_email) is cache-backed with a 60-second TTL. Second request for the same (chat_id, user_email) is a dict lookup. Cache is load-bearing — the MCP middleware calls it on every request to pre-fill the ContextVar for Tier 4.

​
Usage pattern for custom clients

import requests

r = requests.get(
    "http://localhost:8081/system-prompt",
    params={"chat_id": "my-session", "user_email": "user@example.com"},
    headers={"Authorization": f"Bearer {MCP_API_KEY}"},
)
r.raise_for_status()
system_prompt = r.text
public_base_url = r.headers["X-Public-Base-URL"]

# Inject into your LLM call as the system message.
# Use public_base_url when rewriting file links in UI.
EndpointPurpose
GET /skill-list?user_email=...Skills list for sub-agent prompts
GET /skill-mounts?user_email=...Docker bind-mounts for user skills
GET /mcp-infoAvailable tools, required headers, endpoint URL (JSON)
GET /health{"status":"healthy"}