node804-mcp-toolkit
Node804/node804-mcp-toolkit
Shared utilities for building token-efficient, security-conscious MCP servers for IT operations. Provides the cross-cutting concerns every ops MCP needs (permission gating, audit logging, response shaping, TLS config ) so each server in a suite implements them the same way.
Instalação
npx convez-ai Node804/node804-mcp-toolkit1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
node804-mcp-toolkit
Shared utilities for building token-efficient, security-conscious MCP servers for IT operations. Provides the cross-cutting concerns every ops MCP needs — permission gating, audit logging, response shaping, TLS config — so each server in a suite implements them the same way.
Installation
pip install node804-mcp-toolkit
What's in here
| Module | Purpose |
|---|---|
rbac | Hierarchical permission modes (read / standard / full / admin); a gate decorator that only registers tools with the MCP server when the active mode qualifies — ungated tools are invisible to the AI client, not just blocked |
audit | JSON-lines audit logging of every tool call with timing, success/error state, sensitive-key redaction, and long-value truncation |
lean | Pure response-shaping helpers: field whitelisting, internal-key stripping, pagination, server-side pattern filtering |
tls | Verify-by-default TLS config from env vars, with custom CA bundle support for internal PKI |
params | Shared Pydantic parameter types (VerboseFlag, FieldsList, Pagination, Pattern) so every tool uses the same names, defaults, and descriptions |
Quick start
Gate tools by permission mode, with audit logging composed in:
from mcp.server.fastmcp import FastMCP
from node804_mcp_toolkit import Mode, ModeGate, open_sink
mcp = FastMCP("panos-mcp")
# Audit sink: writes JSONL when PANOS_AUDIT_LOG is set, no-op otherwise.
sink = open_sink(env_var="PANOS_AUDIT_LOG")
# Mode comes from env. Missing or invalid values fail safe to read-only.
gate = ModeGate.from_env(env_var="PANOS_MODE", audit_sink=sink)
@gate.tool(mcp, required=Mode.READ)
async def get_security_rules(...): ...
@gate.tool(mcp, required=Mode.ADMIN)
async def commit(...): ... # not registered at all unless PANOS_MODE=admin
Shape responses for token efficiency:
from node804_mcp_toolkit import filter_by_pattern, paginate, strip_keys, whitelist
rules = strip_keys(raw_rules, ["@uuid", "@loc"]) # drop SDK internals
rules = filter_by_pattern(rules, pattern, name_field="name")
rules = paginate(rules, limit=100, offset=0)
rules = whitelist(rules, ["name", "action", "source", "destination"])
Resolve TLS settings from environment (verify-by-default):
from node804_mcp_toolkit import resolve_tls_config
import httpx, os
tls = resolve_tls_config(dict(os.environ), prefix="PRTG")
# PRTG_TLS_VERIFY=false → verify off, with a loud stderr warning
# PRTG_TLS_CA=/path.pem → custom CA bundle for internal PKI
client = httpx.Client(verify=str(tls.ca_path) if tls.ca_bundle else tls.verify)
Design philosophy
- One toolkit, one set of patterns. A user who learns the parameter conventions in any one MCP —
verbose,fields,limit,offset,pattern— gets the same conventions in all of them. - Default deny. Unset or misconfigured mode env vars fall back to read-only. Tools above the active mode are never registered, so the AI client can't even see them.
- Token-lean by default. Responses expose the fields the AI actually reasons over;
verbose=trueopts into the full payload. - Best-effort observability. Audit logging never breaks a tool call — sink failures warn once to stderr and move on.
Used by
node804-panos-mcp— Palo Alto firewall managementnode804-freshservice-mcp— Freshservice ticketing- Planned: PRTG and Veeam MCPs
Development
pip install -e ".[dev]"
pytest
ruff check .
mypy src
Requires Python 3.11+. Fully typed (py.typed included), mypy --strict clean.