Query your Logfire logs and traces

---
name: logfire-query
description: Query and analyze Logfire telemetry data — traces, logs, spans, metrics, summaries, and SQL results. Use this skill when the user asks to "query logfire", "search traces", "find logs", "query data", "search spans", "look up errors in logfire", "get metrics from logfire", "analyze telemetry", "summarize errors", "find root cause", or add Logfire querying capabilities to code. Do not use this skill for direct Logfire UI, browser, live-view, Explore-page, or link-opening requests; use logfire-ui instead. If "show" or "view" wording is ambiguous, ask whether the user wants a UI view or query analysis.
---

# Query Logfire Data

## When to Use This Skill

Invoke this skill when:
- User wants to query traces, logs, spans, or metrics from Logfire
- User wants to search for specific events, errors, or patterns in telemetry data
- User wants to analyze OpenTelemetry data stored in Logfire
- User wants to add programmatic query capabilities to their code
- User asks to "query logfire", "search traces", "find logs", "get metrics", "count", "summarize", "compare", or "find root cause"

## User-Facing Progress

Keep progress updates terse. Do not narrate route classification, local skill instructions, schema selection, or routine query setup. If an update is needed, use one short sentence focused on the action, such as "Querying recent Logfire errors."

## Critical Routing: One Workflow Per Request

Before using any query tool, classify the request.

- Query route: "analyze", "query", "count", "summarize", "compare", "find root cause", "find slowest", "look up errors", "get metrics", or "add query capabilities".
- UI route: "open", "show in browser", "show in Codex", "show in Logfire", "live view", "open Explore", "open the UI", "give me a link", or GUI/browser presentation. Use `logfire-ui`; do not call `query_run` just to make a URL.
- Ambiguous route: prompts like "show recent errors", "view logs", or "show spans" do not specify whether the user wants chat analysis or the Logfire UI. Ask the user to choose query analysis or UI view.
- Combined route: if the user explicitly asks for both analysis and a link, query only for the requested analysis or to identify the requested item, then provide the relevant Logfire link. Do not add UI/browser work unless the user asked for it.

Only query before opening Logfire UI when the user asks to open a specific unknown item that must be found first, such as "find the slowest trace and open it" or "open the latest error trace".

## Two Approaches

| Aspect | MCP `query_run` | REST API `/v1/query` |
|--------|-----------------|----------------------|
| **Best for** | Interactive analysis in Codex | Adding query code to a project |
| **Auth** | OAuth via MCP session | Bearer read token |
| **Setup** | Already configured via plugin | Need a read token |
| **Formats** | JSON rows | JSON, CSV, Apache Arrow |
| **Default window** | Last 30 min | Last 24 hours |
| **Max range** | 14 days | 14 days |
| **Row limit** | Must be in SQL | Default 500, max 10,000 |

## Quick Schema Reference

### `records` table (spans and logs)

Key columns for querying:

| Column | Type | Description |
|--------|------|-------------|
| `start_timestamp` | timestamp (UTC) | When span/log was created |
| `end_timestamp` | timestamp (UTC) | When span/log completed |
| `duration` | double (seconds) | Time between start and end; NULL for logs |
| `trace_id` | string (32 hex) | Unique trace identifier |
| `span_id` | string (16 hex) | Unique span identifier |
| `parent_span_id` | string (16 hex) | Parent span; NULL for root spans |
| `span_name` | string | Low-cardinality label for similar records |
| `message` | string | Human-readable description with arguments filled in |
| `level` | integer | Severity (supports `level = 'error'` string comparison) |
| `kind` | string | `span`, `log`, `span_event`, or `pending_span` |
| `service_name` | string | Service identifier |
| `is_exception` | boolean | Whether an exception was recorded |
| `exception_type` | string | Exception class name |
| `exception_message` | string | Exception message |
| `exception_stacktrace` | string | Full traceback |
| `attributes` | JSON | Structured data; query with `->>'key'` |
| `tags` | string[] | Grouping labels; query with `array_has(tags, 'x')` |
| `http_response_status_code` | integer | HTTP status code |
| `http_method` | string | HTTP method |
| `http_route` | string | HTTP route pattern |
| `otel_status_code` | string | Span status |

### `metrics` table

| Column | Type | Description |
|--------|------|-------------|
| `recorded_timestamp` | timestamp (UTC) | When metric was recorded |
| `metric_name` | string | Metric name |
| `metric_type` | string | Type (gauge, counter, histogram) |
| `unit` | string | Unit of measurement |
| `scalar_value` | double | Metric value |
| `service_name` | string | Service identifier |
| `attributes` | JSON | Metric dimensions |

Full schema: [`references/schema.md`](./references/schema.md)

## SQL Syntax

Logfire uses **Apache DataFusion** (Postgres-like). Key patterns:

```sql
-- Time filtering
WHERE start_timestamp > now() - interval '1 hour'

-- JSON attribute access
WHERE attributes->>'user_id' = '123'
SELECT attributes->>'http.url' as url FROM records

-- Nested JSON
attributes->'request'->>'method'

-- Array filtering
WHERE array_has(tags, 'production')

-- Level filtering (string comparison works)
WHERE level = 'error'

-- Case-insensitive matching
WHERE message ILIKE '%timeout%'

-- Time bucketing for aggregation
SELECT time_bucket(interval '5 minutes', start_timestamp) as bucket,
       count(*) FROM records GROUP BY bucket ORDER BY bucket
```

## MCP Approach (Interactive)

Call the `query_run` MCP tool:
- `query` (required): SQL query string
- `project` (optional): target project (default: user's current project)
- `min_timestamp` / `max_timestamp` (optional): ISO timestamps for time window

Default window is last 30 min. Max range is 14 days. Always include `LIMIT` in SQL.

### Common queries

```sql
-- Recent errors
SELECT start_timestamp, message, exception_type, exception_message
FROM records WHERE is_exception LIMIT 20

-- Slow spans
SELECT span_name, duration, start_timestamp
FROM records WHERE duration > 1.0 ORDER BY duration DESC LIMIT 20

-- Endpoint errors
SELECT start_timestamp, message, http_response_status_code
FROM records WHERE http_route = '/api/users' AND level = 'error' LIMIT 20

-- Full trace
SELECT span_name, message, duration, parent_span_id
FROM records WHERE trace_id = '<id>' ORDER BY start_timestamp

-- Error breakdown by service
SELECT service_name, count(*) as errors
FROM records WHERE is_exception GROUP BY service_name ORDER BY errors DESC
```

## UI Links After Querying

If the user explicitly asks for both analysis and a Logfire link, finish the query analysis first, then use a Logfire link only for the known result:

- For a known `trace_id`, use `project_logfire_link(trace_id=trace_id, project=project, handoff=True)` only when the user asked to open it immediately in the browser. Use `project_logfire_link(trace_id=trace_id, project=project)` for a durable or shareable URL.
- For a project/filter view, use the `logfire-ui` routing rules.
- Do not open the browser unless the user asked to open the link.

For a span-count prompt, provide SQL like this when the user wants an aggregate query or analysis:

```sql
SELECT
  time_bucket(interval '5 minutes', start_timestamp) AS bucket,
  count(*) AS span_count
FROM records
WHERE kind = 'span'
GROUP BY bucket
ORDER BY bucket
LIMIT 200
```

## REST API Approach (Programmatic)

**Endpoint**: `GET https://logfire-api.pydantic.dev/v1/query`

Region variants:
- US: `https://logfire-us.pydantic.dev/v1/query`
- EU: `https://logfire-eu.pydantic.dev/v1/query`

**Auth**: `Authorization: Bearer <read_token>`

**Parameters**:
- `sql` (required): SQL query
- `min_timestamp` / `max_timestamp` (optional): ISO timestamps
- `limit` (optional): row limit (default 500, max 10,000)

**Response formats** (via `Accept` header):
- `application/json` — column-oriented JSON (default)
- `application/json` with `row_oriented=true` param — row-oriented JSON
- `text/csv` — CSV
- `application/vnd.apache.arrow.stream` — Apache Arrow

**Python clients**: `LogfireQueryClient` (sync), `AsyncLogfireQueryClient` (async), `logfire.db_api` (PEP 249 / pandas).

Detailed examples: [`references/client-usage.md`](./references/client-usage.md)

## Query Best Practices

1. **Always LIMIT** — start with 20, increase as needed
2. **Use `min_timestamp`/`max_timestamp` params** for simple time windows instead of SQL `WHERE`
3. **Filter efficiently** — `service_name`, `span_name`, `trace_id`, `is_exception` are fast filters
4. **Use `->>'key'`** for JSON attribute access (returns text); use `->` for nested JSON objects
5. **Avoid `SELECT *`** — select only the columns you need
6. **Max 14-day range** — queries cannot span more than 14 days