Small standalone MCP servers and Skills for agents working on Reticulum and MeshChatX. Go, stdlib-first, stdio transport, safe read-only defaults.
| Server | Purpose |
|---|---|
| rns | RNS manual, zen, forum/GitHub community, rngit, read-only rn* probes |
| reticulum-go | Reticulum-Go docs at reticulum-go.quad4.io: searchable, section-aware, fetch |
| memory | Persistent agent memory: notes, people, RNS destinations, tasks, fuzzy recall |
| meshchatx | MeshChatX docs, scaffolding, god-file detection, split plans, surface snapshots, opt-in GitHub issue tools (env-token gated) |
| context | Token-efficient code access: outlines, symbol reads, repo maps |
| agents | Repo skills, conventions, module ownership, file tree |
| no-slop | Anti-slop linter for text, diffs, files, directories |
| bug-hunter | Hotspots, TOCTOU, attack surface, complexity, regression mining |
| ci-security | GitHub Actions + Dockerfile scanning, action pinning, YAML lint |
| i18n | Locale keys, coverage, usage, hardcoded strings |
| workspace | Taskfile list/run, git status/log/diff, test mapping and runs |
| lxmf | Local Reticulum state: config, storage, known destinations |
| lxmfy | LXMFy docs, bot scaffolding, diagnostics, and test guidance |
| micron | Micron parsing, linting, rendering, extraction, search, templates, and syntax reference |
| gateway | Multiplexes all servers behind tools/tool_schema/invoke. One config entry, tiny tool surface. |
| mediawiki | MediaWiki Action API on one configured wiki: search, pages, backlinks, categories; solves Anubis PoW natively, Cloudflare via optional FlareSolverr |
git clone git@github.com:Quad4-Software/ai.git
cd ai
make all
make gosec
make install
make install prints a single client mcp.json entry for gateway plus a generated quad4-mcp.json for the gateway to consume. Copy the gateway entry into your MCP client config, then place the gateway config at ~/.config/mcp/quad4-mcp.json. The gateway lazily spawns the other servers behind one compact tool surface.
make mcp-config # regenerate dist/mcp.json and dist/quad4-mcp.json
make server-json # generate registry server.json files per server
make inspector # smoke test every built server for tools/list
Prebuilt binaries are also attached to each GitHub release.
Skills live in .agents/skills/ and follow the Agent Skills specification. Install them into your agent with the skills.sh CLI:
npx skills add Quad4-Software/ai
Useful flags:
npx skills add Quad4-Software/ai --list # preview available skills
npx skills add Quad4-Software/ai --skill micron # install only one skill
npx skills add Quad4-Software/ai -g # global install
Skill groups are defined in skills.sh.json, for example MCP toolkit (build, test, security, release, style), Reticulum (mesh networking, LXMF, tooling, interface operation), self-hosted stack, security, and SurrealDB.
Point your client at gateway instead of the individual servers. It spawns each configured command server lazily, exposes a compact tool index, and fetches full schemas only on demand. External servers in the same config (docker-based, etc.) are proxied too.
To share one gateway across every client window, use shared daemon mode: register the gateway with args ["--attach"], which bridges each window's stdio to a single long-lived daemon on an owner-only unix socket (auto-started on first connect). See mcp/gateway/README.md.
mcp-servers.json is the single source of truth for every server. The scripts/gen-configs.py tool reads it to emit the client mcp.json, the gateway quad4-mcp.json, and per-server registry server.json files. The mcp-servers.json list also drives the skills.sh descriptions.
- Newline-delimited JSON-RPC 2.0 over stdio. MCP protocol 2025-11-25.
- Malformed input returns a parse error, never a crash.
- A panicking tool handler returns isError content. The server lives.
- Tool failures return error text naming the required arguments.
- Filesystem access is path-jailed to the repo root. Secrets redacted.
- No arbitrary command execution anywhere.
Each server is one binary. Add to ~/.config/mcp/mcp.json under mcpServers, for example:
"rns": {"command": "/path/ai/mcp/rns/rns"},
"context": {"command": "/path/ai/mcp/context/context",
"env": {"MCP_REPO_ROOT": "/path/to/repo"}}
Use memory to persist destinations, learned facts, and todo items.
License: 0BSD