Langfuse
Give the Sazabi agent tools to inspect LLM traces, observations, metrics, scores, comments, and prompts from your Langfuse project.
All Langfuse integration capabilities
About
Langfuse is an MCP connector that gives the Sazabi agent tools to inspect LLM traces, observations, metrics, scores, comments, and prompts. Langfuse hosts an MCP server over Streamable HTTP at /api/public/mcp, region-scoped by Langfuse Cloud host — US (https://us.cloud.langfuse.com), EU (https://cloud.langfuse.com), or JP (https://jp.cloud.langfuse.com). Sazabi authenticates using HTTP Basic Auth built from a Langfuse project API key pair (pk-lf-... / sk-lf-...), stored securely and sent in the Authorization header.
What the agent can do:
- Query traces and observations to investigate LLM application behavior
- Retrieve metrics, scores, and evaluation results
- Read comments attached to traces and observations
- Fetch managed prompts from Prompt Management
- Scope the connection to your Langfuse Cloud region (US, EU, or JP)
Prerequisites
- A Langfuse Cloud account (or self-hosted deployment via the custom connector)
- A Langfuse project API key pair (Public
pk-lf-...and Secretsk-lf-...) from Project settings > API keys
Set up in the dashboard
Configure connectors in the dashboard under Settings > MCP connectors.
Open the connector
Find Langfuse in Settings > MCP connectors.
Add your key
Select your Langfuse Cloud region (US, EU, or JP). In Langfuse, open Project settings > API keys and create a new key pair. Base64-encode publicKey:secretKey and enter it as Basic <encoded>. Sazabi sends this value as the Authorization header to Langfuse's hosted MCP endpoint.
Set up with the CLI
Langfuse authenticates with Langfuse project API keys. With the Sazabi CLI installed and authenticated (see sazabi mcp-connectors reference), connect it in one command, passing the credential as a header:
sazabi mcp-connectors connect langfuse --header "Authorization=Basic <encoded>"The CLI connects to the MCP server at once and prints the connection with the tools it discovered; a credential the server rejects is reported, and the connection is left needing attention. Use NAME=@<file> to read a value from a file, or --header-file <file.json> for a JSON object of header names and values. To rotate the stored credentials later, use sazabi mcp-connectors update-credentials <connection-id>; find the connection ID with sazabi mcp-connectors list.
Verify
Open the connector's connection page under Settings > MCP connectors and confirm Sazabi lists the tools it discovered. Then, in a thread, ask the agent to use one of those tools and confirm it runs. If no tools appear, confirm the connection is authorized (OAuth) or the key/header is correct (token auth).
Troubleshooting
No tools discovered — Confirm the Authorization value is correctly formatted as Basic <base64(pk-lf-...:sk-lf-...)> and that you selected the correct region; re-enter from Settings > MCP connectors.
Connection shows needs attention — The authorization or key may have expired or been revoked; reconnect from Settings > MCP connectors.