---

name: langfuse version: 1.0.1 description: Debug AI traces, find exceptions, analyze sessions, and manage prompts via Langfuse MCP. Also handles MCP setup and configuration. metadata: short-description: Langfuse observability via MCP compatibility: claude-code, codex-cli

Langfuse Skill

Debug your AI systems through Langfuse observability.

Triggers: langfuse, traces, debug AI, find exceptions, set up langfuse, what went wrong, why is it slow, datasets, evaluation sets

Setup

Step 1: Get credentials from https://cloud.langfuse.com → Settings → API Keys

If self-hosted, use your instance URL for LANGFUSE_HOST and create keys there.

Step 2: Install MCP (pick one):

Copy# Claude Code (project-scoped, shared via .mcp.json)
claude mcp add \
  --scope project \
  --env LANGFUSE_PUBLIC_KEY=pk-... \
  --env LANGFUSE_SECRET_KEY=sk-... \
  --env LANGFUSE_HOST=https://cloud.langfuse.com \
  langfuse -- uvx --python 3.11 langfuse-mcp

# Codex CLI (user-scoped, stored in ~/.codex/config.toml)
codex mcp add langfuse \
  --env LANGFUSE_PUBLIC_KEY=pk-... \
  --env LANGFUSE_SECRET_KEY=sk-... \
  --env LANGFUSE_HOST=https://cloud.langfuse.com \
  -- uvx --python 3.11 langfuse-mcp

Step 3: Restart CLI, verify with /mcp (Claude) or codex mcp list (Codex)

Step 4: Test: fetch_traces(age=60)

Read-Only Mode

For safer observability without risk of modifying prompts or datasets, enable read-only mode:

Copy# CLI flag
langfuse-mcp --read-only

# Or environment variable
LANGFUSE_MCP_READ_ONLY=true

This disables write tools: create_text_prompt, create_chat_prompt, update_prompt_labels, create_dataset, create_dataset_item, delete_dataset_item.

For manual .mcp.json setup or troubleshooting, see references/setup.md.

---

Playbooks

"Where are the errors?"

Copyfind_exceptions(age=1440, group_by="file")

→ Shows error counts by file. Pick the worst offender.

Copyfind_exceptions_in_file(filepath="src/ai/chat.py", age=1440)

→ Lists specific exceptions. Grab a trace_id.

Copyget_exception_details(trace_id="...")

→ Full stacktrace and context.

---

"What happened in this interaction?"

Copyfetch_traces(age=60, user_id="...")

→ Find the trace. Note the trace_id.

If you don't know the user_id, start with:

Copyfetch_traces(age=60)

Copyfetch_trace(trace_id="...", include_observations=true)

→ See all LLM calls in the trace.

Copyfetch_observation(observation_id="...")

→ Inspect a specific generation's input/output.

---

"Why is it slow?"

Copyfetch_observations(age=60, type="GENERATION")

→ Find recent LLM calls. Look for high latency.

Copyfetch_observation(observation_id="...")

→ Check token counts, model, timing.

---

"What's this user experiencing?"

Copyget_user_sessions(user_id="...", age=1440)

→ List their sessions.

Copyget_session_details(session_id="...")

→ See all traces in the session.

---

"Manage datasets"

Copylist_datasets()

→ See all datasets.

Copyget_dataset(name="evaluation-set-v1")

→ Get dataset details.

Copylist_dataset_items(dataset_name="evaluation-set-v1", page=1, limit=10)

→ Browse items in the dataset.

Copycreate_dataset(name="qa-test-cases", description="QA evaluation set")

→ Create a new dataset.

Copycreate_dataset_item(
  dataset_name="qa-test-cases",
  input={"question": "What is 2+2?"},
  expected_output={"answer": "4"}
)

→ Add test cases.

---

"Manage prompts"

Copylist_prompts()

→ See all prompts with labels.

Copyget_prompt(name="...", label="production")

→ Fetch current production version.

Copycreate_text_prompt(name="...", prompt="...", labels=["staging"])

→ Create new version in staging.

Copyupdate_prompt_labels(name="...", version=N, labels=["production"])

→ Promote to production. (Rollback = re-apply label to older version)

---

Quick Reference

age = minutes to look back (max 10080 = 7 days)

---

References

• references/tool-reference.md — Full parameter docs, filter semantics, response schemas
• references/setup.md — Manual setup, troubleshooting, advanced configuration