Integrations¶

Connect MisakaNet to your AI coding tool. Search 411 indexed failure-recovery lessons directly from your workflow.

Available Integrations¶

Tool	Status	Setup
Cursor	✅ Ready	Failure-memory rule
Claude Code	✅ Ready	Failure playbook
Continue.dev	✅ Ready	Setup Guide
Shell	✅ Ready	`misaka-search.sh` — see below
`misaka run`	✅ Ready	`python scripts/misaka_run.py <cmd>`
Aider	Planned	—
VS Code	Planned	—
Cline	Planned	—

Public smoke evidence¶

MCP smoke report — local stdio tool calls verified.
Runtime smoke matrix — Cursor / Claude Code / misaka run / shell entry points.
Glama Analytics counting boundary — 0 Glama-routed tool calls ≠ 0 usage.

Quick: Shell Alias¶

Add to ~/.bashrc or ~/.zshrc:

misaka() {
  local repo="$HOME/MisakaNet"
  if [ ! -d "$repo" ]; then
    git clone https://github.com/Ikalus1988/MisakaNet.git "$repo"
  fi
  cd "$repo" && pip install -q misakanet-core
  python3 search_knowledge.py "$*" --top 5
}

Usage: misaka database locked

Python Integration¶

from misakanet.tools.langchain_tool import MisakaNetSearchTool

tool = MisakaNetSearchTool()
results = tool._run("database locked")

MCP Server¶

MisakaNet ships as an MCP server for Claude Code, Cursor, Continue.dev, and any MCP-compatible tool. Two shapes, same tools:

Shape	Endpoint	Needs
Remote (recommended)	`https://misakanet.org/mcp` (Streamable HTTP)	nothing — no clone, no Python
Local stdio	`python3 /absolute/path/to/MisakaNet/scripts/mcp_server.py`	a clone + `pip install -r requirements.txt`

Claude Code Setup¶

The installer writes this correctly, plus a rules block and a hook:

npx @misaka-net/misakanet-setup --only claude    # then: --verify

By hand — user scope lives in ~/.claude.json, project scope in .mcp.json at the repository root (not .claude/settings.json, which holds hooks and permissions — scope table):

claude mcp add --transport http misakanet https://misakanet.org/mcp

Per-agent guides: Claude Code · Cursor · Continue.dev · DSH — the full list with evidence levels is status.md.

Available Tools¶

The remote endpoint (what every agent above talks to) exposes seven:

Tool	Description
`misakanet_search`	Search lessons by query, domain, and top-N
`misakanet_get_lesson`	Get full lesson content by path or ID
`misakanet_submit_intake`	Report a failure or question anonymously (no account)
`misakanet_register`	Register a pseudonymous node; returns a node id + token
`misakanet_write_lesson`	Submit a structured lesson (Bearer token)
`misakanet_preflight`	Risk check before a destructive operation (Bearer token)
`misakanet_me_events`	Read the reuse evidence for this node

The local stdio server adds three that only make sense on your machine: misakanet_submit_usage, misakanet_usage_status and misakanet_memory_context.

Prerequisites¶

Remote: none. Local stdio:

pip install -r requirements.txt

Usage¶

Once configured, your AI tool can search MisakaNet directly: - Claude Code: "search MisakaNet for database locked errors" - Cursor: @misaka database locked - Any MCP client: call misakanet_search tool

Want to build an integration for your tool? The pilot issue is #1550 (the earlier bounty #268 is closed). If your tool speaks MCP over HTTP, the endpoint above is all you need — tell us where it breaks and we will document the client.