Navi MCP Server

This repository is the result of a full audit and rebuild: the server's tool surface was corrected against an authoritative recursivenavi --helpcapture, and all skills were corrected to match the server + CLI and restructured for progressive disclosure.

server/ server.py — the MCP server (19 tools + resources) skills/ the 13 corrected skills, in NAVI_SKILL_DIR layout (<skill>/SKILL.md, plus references/ on the denser ones) dist/ the same 13 skills packaged as .skill files tools/ navi_mcp_config.py — auto-detects paths, emits the install config docs/ audit framework, gap ledger, verified findings, help-crawler INSTALL.md step-by-step install for Claude Desktop README.md this file

navi(router) ·navi-core·navi-mcp·navi-troubleshooting·navi-acr·navi-export·navi-scan·navi-was·navi-action·navi-mail·navi-remote-exec·navi-explore·navi-enrich

EachSKILL.mdis under 500 lines. Deep material (full schema, exhaustive selector catalog, long worked examples) lives inreferences/*.mdand is pulled on demand.

The server shells out to thenavibinary and reads the localnavi.db. It doesnotmanage API keys — set those out-of-band withnavi config keysfirst (seeskills/navi-core).

Launch it over stdio (for Claude Desktop / Claude Code) or HTTP:

python server/server.py # stdio (default); waits for a client python server/server.py --http # streamable HTTP on :8000

NAVI_EMAILandNAVI_REMOTE_CODE_EXECUTIONare deliberate, separate opt-ins stacked on the master write gate: enabling writes alone doesnotenable email or remote execution.navi_action_mailneeds SMTP set vianavi config smtp;navi_action_pushneeds SSH creds vianavi config ssh(both out-of-band). The config helper exposes matching flags:--allow-emailand--allow-remote-code-execution(each used alongside--allow-writes).

PointNAVI_SKILL_DIRat this repo'sskills/folder (notdist/— the server reads unpacked folders, not.skillzips).

python tools/navi_mcp_config.py # print the mcpServers JSON python tools/navi_mcp_config.py --write # merge it into your Claude Desktop config (backs up first)

The launched server entry isserver/server.py(use anabsolutepath in the config — Claude Desktop won't have your shell'sPATH). After editing the config, fully quit and reopen Claude Desktop, then readnavi://workdirto confirm it connected.

Read tools (navi_explore_data,navi_explore_info,navi_explore_querySELECT,navi_export,navi_explore_apiGET, scan read views, …) run freely.Platform-write tools require bothNAVI_MCP_ALLOW_WRITES=1andconfirm=Trueand are meant to be narrated to the user before invocation:navi_enrich_tag,navi_enrich_acr,navi_enrich_add,navi_scan(create/start/stop/pause/resume),navi_was(scan/start/upload),navi_action_delete,navi_action_rotate,navi_action_cancel,navi_config(kind="url"), andnavi_explore_apiPOST/PUT.

Double-gated toolsneed a second capability env var on top of the write gate, plusconfirm=True:

- navi_action_mail— email; requiresNAVI_MCP_ALLOW_WRITES=1andNAVI_EMAIL=1. Harness:skills/navi-mail.
- navi_action_push— remote command/file execution over SSH; requiresNAVI_MCP_ALLOW_WRITES=1andNAVI_REMOTE_CODE_EXECUTION=1. Harness:skills/navi-remote-exec.

- navi://schema/{table}— live column definitions for a navi.db table
- navi://workdir— workdir, write-gate status, binary, call budget, and navi.db freshness
- navi://skill/{name}— load a skill (router/core/mcp/…); lists its references
- navi://skill/{name}/{ref}— load a bundled reference (e.g.navi://skill/core/schema)

Plus thenavi_workflowprompt, which injects the router skill.

navi exports can run for tens of minutes on large tenants — past the MCP host's ~4-minute tool-call ceiling. The server enforces a call budget (~220s) and returns a clean error naming the CLI command to run instead. Foundational syncs (navi config update full) remain intentionally CLI-only. Remote command execution and email (navi_action_push/navi_action_mail) are now exposed as tools, but each is double-gated (see above). Seeskills/navi-coreandskills/navi-troubleshooting.

The files indist/are packaged for installing in Claude.ai / Claude Cowork / Claude Code as skills. (The MCP server uses the unpackedskills/folders viaNAVI_SKILL_DIR; the two are the same content in two delivery formats.)

server.pycompiles cleanly and every tool is annotated. It hasnotbeen runtime-tested against a live Tenable tenant. Before relying on it, validate the root-cause fix with one live read, e.g.navi_explore_data(subcommand="cve", cve="CVE-2021-44228"). The tool annotations require a recentmcpSDK.

Seedocs/verified-findings.mdfor the per-bug inventory anddocs/gap-ledger.mdfor the full audit trail.

Civilian situational awareness for AI deployments — real-time risk dashboards, multi-source threat correlation, anomaly detection, and automated alerting for critical infrastructure and enterprise AI systems.

Monitor WAF events, analyze attacks, tune rules and whitelist IPs for OWASP ModSecurity CRS via Docker

Provides a unified interface to AWS services for security investigations and incident response.

Network reconnaissance and security scanning with port scanning, DNS analysis, and vulnerability assessment

Self-hosted MCP governance runtime in Rust — audit trail, policy enforcement, and cost controls for MCP servers.

Integrate with Tailscale's CLI and API for automated network management and monitoring.

Secure Zero-Trust SSH Gateway for AI Agents. A Go-based Model Context Protocol (MCP) server featuring runtime Regex Command Firewalls and multi-host isolation.

Secure every MCP server with one governed gateway. Give each AI agent its own scoped MCP access, contain credentials at the gateway, and audit every MCP tool call without wiring agents directly to each server.

Read-only MCP server inside a self-hosted homelab dashboard — explore hosts, Docker containers, GPU/VRAM, systemd services, AI models, alerts and disk.