ai-raccoon-live-probe

Probe a running AiRaccoon MCP server via HTTP, SQLite, and CLI diagnostics.

2|Updated Aug 2, 2026
One-click install
npx skills add https://github.com/Arasz/ai-raccoon --skill ai-raccoon-live-probe-arasz
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ai-raccoon-live-probe
Source: https://github.com/Arasz/ai-raccoon/tree/main/.ai-badger/skills/learned/uncategorized/ai-raccoon-live-probe
Command: npx skills add https://github.com/Arasz/ai-raccoon --skill ai-raccoon-live-probe-arasz

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Investigating a live AiRaccoon memory server is error-prone: the data root is hidden, MCP responses arrive as SSE frames, settings state is invisible, and sqlite reads of vec0 shadow tables are easy to misinterpret. This Skill captures the exact mechanics for health-checking, querying, and safely inspecting a running server without disturbing the user's instance. ## Core Features & Use Cases - Live server discovery: Locate the data root, bearer token, and bound port of a running ai-raccoon process using pgrep, lsof, and server logs. - Health check protocol: Run an ordered sequence of process checks, MCP initialize, doctor, memory_stats, and memory_search to verify end-to-end retrieval. - Wire protocol decoding: Parse SSE-framed MCP responses, double-encoded tools/call payloads, and nested query-guard warning annotations correctly. - Safe bank inspection: Read memory.db settings and vec0 shadow tables while the server runs, and spin up isolated scratch servers on port 0 for A/B settings tests. - Use Case: When a manual checklist item fails or retrieval quality looks wrong, use this Skill to confirm whether fusion, noise, or query-guard settings actually engaged via their telemetry gauges before blaming the binary. ## Quick Start Probe the running ai-raccoon server to confirm its health, project ID, and current settings state.

Frequently Asked Questions about ai-raccoon-live-probe

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I find the data root of a running ai-raccoon server?

Find the process with pgrep -fl "ai-raccoon", then run lsof -p <pid> and grep for .db to reveal the memory.db path, whose parent directory is the data root. The bearer token sits at <data-root>/mcp-token, and the macOS default is ~/.ai-raccoon.

How do I health check an ai-raccoon MCP server?

Run checks in order: verify the process and listening port 7721, POST an MCP initialize request, run ai-raccoon doctor for schema status, then call memory_stats and a short memory_search to confirm hybrid retrieval works end-to-end.

Why does parsing MCP responses from ai-raccoon fail with JSONDecodeError?

Every response is an SSE frame, even non-streaming ones, so a parser expecting bare JSON fails. Collect lines starting with data:, take the last one, and json.loads it; tools/call payloads are additionally JSON strings inside content[0].text requiring a second parse.

Can I read the ai-raccoon memory.db with sqlite3 while the server runs?

Yes, a plain sqlite3 open works; use PRAGMA query_only=1 or a ?mode=ro URI for safety on someone else's bank. Note that vec0 virtual tables are unreadable to system sqlite, so read the vec_entries_vector_chunks00 shadow table instead.

Why does ai-raccoon doctor report a shape mismatch that never resolves?

On pre-1.33.8 binaries the sync_tombstones repair only ran when the schema digest mismatched, leaving a permanent mismatch. Since 1.33.8 the repair runs as ladder step v11 on any older bank, so starting the server once and re-running doctor clears it.

How do I test ai-raccoon settings without touching the live server?

Start a scratch server with ai-raccoon --data-root <scratch-dir> serve --port 0 and read the bound port from its own log output. Never bind port 7721 or use --restart, since the user's live server owns it.