gcplogging-mcp

MCP server for bulk downloading Google Cloud Logging entries to local files for offline analysis.

Tools

Tool	Description
list_projects	List configured GCP projects
list_logs	List available log names in a project
download_logs	Fetch log entries and write to a local JSONL file

download_logs usage

project:     Config key, e.g. "production"
filter:      Cloud Logging filter, e.g. 'resource.type="cloud_run_revision" AND severity>=ERROR'
start_time:  ISO 8601, e.g. "2024-01-20T00:00:00Z"
end_time:    ISO 8601 (optional, defaults to now)
output_file: Absolute path (optional, defaults to /tmp/gcplogging-{project}-{timestamp}.jsonl)

Returns { "file": "/tmp/...", "entry_count": 42301, "bytes_written": 15234567, "filter": "..." }.

Output is JSONL — one JSON object per line with timestamp, severity, log_name, resource, payload, labels.

Setup

1. Install dependencies

cd gcplogging-mcp
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt

2. Create config

mkdir -p ~/.config/gcplogging-mcp
cp example-config.toml ~/.config/gcplogging-mcp/config.toml
chmod 600 ~/.config/gcplogging-mcp/config.toml
# edit the file with your project IDs and credentials

If credentials_file is omitted, Application Default Credentials are used (gcloud auth application-default login). The service account needs the roles/logging.viewer role.

3. Create wrapper script

#!/bin/sh
export GCPLOGGING_MCP_CONFIG="$HOME/.config/gcplogging-mcp/config.toml"
cd /path/to/gcplogging-mcp
exec .venv/bin/python3 -m gcplogging_mcp.server

4. Register with your MCP client

Point your MCP client at the wrapper script. For a TOML-based config:

[mcp_servers.gcplogging]
command = "/path/to/wrapper/gcplogging-mcp"
startup_timeout_sec = 30.0
tool_timeout_sec = 300.0

Filter examples

# Cloud Run service logs
resource.type="cloud_run_revision" AND resource.labels.service_name="my-service"

# Errors only
resource.type="cloud_run_revision" AND severity>=ERROR

# Load balancer access logs
resource.type="http_load_balancer"

# Text search
resource.type="cloud_run_revision" AND textPayload:"Exception"