# Export a log file to OpenTelemetry traces
exp2span export <log-file>

# Validate a log file format
exp2span validate <log-file>

# Show log metadata and statistics
exp2span info <log-file>

# With verbose output
exp2span --verbose export <log-file>

# With dry-run mode (no actual export)
exp2span export --dry-run <log-file>

# Export to custom OTLP endpoint
exp2span export <log-file> --otlp-endpoint http://localhost:4318

# Export with gRPC protocol
exp2span export <log-file> --otlp-protocol grpc --otlp-endpoint http://localhost:4317

# Export with authentication headers
exp2span export <log-file> --otlp-header "Authorization=Bearer $TOKEN"

# Export to production collector
exp2span export <log-file> \
  --otlp-endpoint https://otel-collector.example.com:4318 \
  --otlp-protocol http \
  --otlp-service-name my-app \
  --otlp-header "Authorization=Bearer $TOKEN" \
  --batch-size 200 \
  --otlp-timeout-secs 60

MCP Semantic Conventions#

This project implements the OpenTelemetry semantic conventions for Model Context Protocol (MCP):

PR: #2083
Status: Merged (Jan 12, 2026)
Author: lmolkova
File: .test-agent/sem/docs/gen-ai/mcp.md

Key Attributes#

Client Spans#

mcp.session.id - Session identifier
mcp.method.name - Request/notification method (e.g., tools/call, initialize)
mcp.protocol.version - MCP version (e.g., 2025-06-18)
gen_ai.operation.name - Operation type (e.g., execute_tool)
gen_ai.tool.name - Tool name
jsonrpc.protocol.version - JSON-RPC version (default: 2.0)
network.transport - Transport protocol (tcp, quic, pipe)
network.protocol.name - OSI application layer (e.g., http, websocket)

Metrics#

mcp.client.operation.duration - Histogram of operation durations
mcp.server.operation.duration - Server-side operation duration

Well-Known MCP Methods#

Common methods from the specification:

tools/call - Execute a tool
tools/list - List available tools
initialize - Initialize MCP client
resources/list - List available resources
resources/read - Read a resource

Implementation#

The parser extracts:

Session ID from header metadata
Assistant/User messages with timestamps
Tool usage information (when available)
Thinking sections for context analysis

Each assistant message is converted to an OpenTelemetry span with appropriate MCP attributes.

OTLP Export#

The exporter supports real-time export to OpenTelemetry-compatible backends:

Protocols:

HTTP (default) - Exports to http://localhost:4318 or custom endpoint
gRPC - Exports to http://localhost:4317 or custom endpoint

Configuration:

--otlp-endpoint - OTLP collector URL
--otlp-protocol - http or grpc
--otlp-service-name - Service name for traces (default: exp2span)
--otlp-header - Custom headers (e.g., Authorization=Bearer $TOKEN)
--batch-size - Number of spans to batch (default: 100)
--otlp-timeout-secs - Request timeout (default: 30)
--otlp-max-retries - Maximum retry attempts (default: 3)
--otlp-retry-delay-ms - Initial retry delay (default: 100)

Supported Collectors:

Jaeger (http://localhost:4318 for HTTP, http://localhost:4317 for gRPC)
Tempo (any OTLP endpoint)
Zipkin (via OTLP)
Grafana Tempo (via OTLP)
Any OpenTelemetry-compatible collector

Dependencies#

opentelemetry - Core tracing API
opentelemetry_sdk - SDK implementation
opentelemetry-otlp - OTLP exporter with HTTP and gRPC support
jiff - Datetime parsing and handling with DST support
regex - Log parsing
serde_json - JSON parsing

References#

MCP Semantic Conventions (PR #2083) - Official OpenTelemetry semantic conventions for Model Context Protocol

Clone this repository

exp2span#

Usage#