pawsift_set_target_packageLABSCO SUMMARY
Logcat cut down to what fits in a reply.
Raw logcat is the wrong shape for an assistant: it arrives faster than anyone reads and repeats itself constantly. This sits in between, keeping a session's worth in SQLite so the questions you actually have โ what errored, what did this tag say, what happened around that line โ are queries rather than scrollbacks.
What it is
A Go binary that polls logcat into a local SQLite store and exposes it as queries. It detects app restarts and rotates sessions on its own, folds repeated lines together, and caps how many lines any answer returns so a log dump cannot swallow the conversation.
What you get
- Logs filtered by level, tag and search term, with a line limit on every answer
- Errors pre-aggregated into a summary rather than returned raw
- The active tags in the current session, listed
- A window of lines either side of one event
- Search across the whole history, not just the current session
- Watcher health, session id and backlog from a status call
- Retention limits adjustable while it runs, and both the local store and the device buffer clearable
Requirements
An Android device or emulator whose logcat this machine can read, and the pawsift binary โ the install script fetches the right build for your OS and puts it in ~/.local/bin.
Setup effort
One command โ curl -fsSL https://raw.githubusercontent.com/dolphprefect/pawsift-mcp/main/install.sh | sh
USE CASES
Finding out what crashed
Errors come back already grouped rather than as a wall of stack traces, and once you have the one that matters you pull the lines either side of it to see what led in.
Following one component's chatter
List which tags are actually active in this session first, then query just that tag โ which is faster than guessing the tag name a library logs under.
Watching a single app while you drive it
Point the ingestor at one package so the store fills with that app's output, then query by level as you reproduce the problem by hand.
Finding an event you saw two runs ago
Search the whole history rather than the current session, so a message that appeared before the last app restart is still reachable.
Checking the capture itself is healthy
When answers look thin, the status call reports whether the watcher is alive, which session it is on and how far behind it is โ the difference between a quiet app and a stopped ingestor.
Keeping the store from growing forever
Raise or lower the retention limits without restarting, and clear the local store or the device buffer when you want a clean run.
9 tools
Listed in project docs. Your client may show a different list.
pawsift_get_statuspawsift_get_error_summarypawsift_get_tag_summarypawsift_query_logspawsift_get_log_contextpawsift_search_logspawsift_clear_logspawsift_set_retention_policyBEFORE YOU INSTALL
What to know first
Tools you get
9 tools named in the project's documentation.
Where it runs
Both documented clients run the server on your machine.
INSTALLATION
Connect your client
Server setup
curl -fsSL https://raw.githubusercontent.com/dolphprefect/pawsift-mcp/main/install.sh | shMCP Clients
Gemini CLI, Claude Code
{
"mcpServers": {
"pawsift": {
"command": "/home/<YOUR_USERNAME>/.local/bin/pawsift"
}
}
}