Skip to main content
Glama

Fairwire IT and network agents

Thirty-one small MCP agents for IT teams. Each one owns one job, such as reviewing firewall rules, checking backup results, reading sign-in logs or reviewing the cloud bill, and cannot act outside it. They read the files your IT team already exports from firewalls, controllers, identity providers, device management, backup tools, service desks and cloud consoles, write reports on your machine, and draft notes. Every change to a live device, firewall, DNS zone, directory or cloud account stays one named person's click.

Part of Mythos Machines (Fairwire on the site). The software is free (MIT); you pay a fixed price for the setup on your files.

MIT licence · Node 20 or newer · local files only · no credentials · no dependencies · names and addresses masked for the model

Also in: Deutsch · Français · Español · Português (Brasil) · Italiano · 日本語 · 简体中文 · 한국어 (translated by Claude; native review pending)

Try one

npx -y fairwire-firewall-rules demo     # firewall rule review on fictional rules
npx -y fairwire-sign-in-logs demo      # sign-in log review on fictional sign-ins

Every agent has a demo that runs on fictional files, an eval that checks it against its own test cases, and an MCP server (npx -y <agent> mcp --root <folder>) for Claude Desktop, Claude Code, Cursor and other MCP hosts. Each agent's README lists its checks, settings, tools and the columns it reads.

Related MCP server: fairshop-agents

The agents

Network and edge

Agent

What it does

Waits for a person before

fairwire-firewall-rules

Firewall rules: any-to-any, risky ports from the internet, shadowed, duplicate, expired, unused, ownerless and unlogged rules

Submitting a change request

fairwire-network-traffic

NetFlow, IPFIX or sFlow: risky ports reached from outside, cleartext protocols, out-of-hours uploads, scans, top talkers

Submitting a change request

fairwire-vpn-access

VPN logs: leavers connecting, two countries at once, failed bursts, shared accounts, long sessions, old clients

Sending an access note

fairwire-dns-queries

DNS query logs: tunnelling signals, outside resolvers, blocked lookups, NXDOMAIN-heavy clients, failing and chatty names

Sending a network note

fairwire-dns-records

DNS zones: dangling cloud CNAMEs (takeover risk), CNAME conflicts, private or foreign addresses, TTLs, wildcards

Submitting a change request

fairwire-wifi-health

Wi-Fi controllers: rogues, WEP/TKIP/open networks, offline and outdated APs, bad channels, weak and sticky clients

Submitting a change request

fairwire-switch-ports

Switch ports and VLANs: trunks with every VLAN, VLAN 1, half duplex, CRC errors, unused ports, unused VLANs

Submitting a change request

fairwire-dhcp-leases

DHCP: rogue servers, conflicts, near-full scopes, misplaced reservations, unknown devices

Sending a network note

fairwire-call-quality

Call quality: low MOS, loss, jitter, one-way audio, poor sites, trunks and hours, dropped calls

Sending a provider ticket

Identity, mail and admins

Agent

What it does

Waits for a person before

fairwire-sign-in-logs

Sign-in logs: password spraying, success after failures, impossible travel, legacy protocols, service accounts used by people

Sending a security note

fairwire-mfa-coverage

MFA coverage: admins without phishing-resistant MFA, no MFA, exclusions, SMS-only, shared mailboxes

Sending an identity note

fairwire-admin-roles

Admin roles: too many top admins, guests, no MFA, break-glass use, standing roles, admins with mail

Sending an identity note

fairwire-directory-hygiene

Directory hygiene: password flags, group loops, stale and unsupported computers, ownerless and empty groups

Sending a clean-up list

fairwire-email-gateway

Email gateway: released threats, allow-list abuse, outbound spam, late pull-backs, data-loss hits

Sending a security note

Devices and software

Agent

What it does

Waits for a person before

fairwire-asset-inventory

CMDB against what is seen: retired but seen, unknown devices, duplicate serials, ghosts, warranties

Sending a CMDB fix list

fairwire-patch-status

Patch status: overdue critical updates, unsupported OS, failures, silent devices, pending restarts

Sending a patch note

fairwire-endpoint-protection

Endpoint protection: no agent, protection or tamper off, passive mode, stale signatures, old high alerts

Sending a security note

fairwire-mobile-devices

Mobile devices: jailbroken, leavers' devices, no passcode or encryption, silent, old OS

Sending a device note

fairwire-software-licenses

Software against licences: prohibited, over-deployed, outdated, unapproved, shelfware

Sending a true-up note

fairwire-print-fleet

Print fleet: offline, idle, costly per page, colour-heavy, leases ending

Sending a consolidation note

Operations and resilience

Agent

What it does

Waits for a person before

fairwire-backup-jobs

Backup results: uncovered critical systems, missed RPOs, repeat failures, size drops, no offsite copy

Sending a backup note

fairwire-restore-tests

Restore tests: never tested, failed, overdue, RTO and RPO missed, no sign-off; a test calendar

Sending a test plan

fairwire-uptime-incidents

Uptime and incidents: below target, missing reviews, repeat failures, slow response, error budgets

Sending a review note

fairwire-change-calendar

Change calendar: freeze clashes, no approval, overlaps, no back-out plan, risky timing

Sending a note to the change board

fairwire-ticket-sla

Ticket SLAs: breaches by priority, unassigned, parked and reopened tickets, team breach rates

Sending a status note

fairwire-log-retention

Logging: retention under policy, silent sources, systems not logging, volume jumps

Sending a logging note

IT spend and AI usage

Agent

What it does

Waits for a person before

fairwire-cloud-costs

AWS, Azure and Google Cloud bills: regions, jumps, untagged spend, commitments, always-on dev, egress

Sending an owner note

fairwire-contract-renewals

IT contracts: notice deadlines, uplifts, overlaps, spend without or after a contract; a deadline calendar

Sending a notice letter

fairwire-telecom-bills

Telecom bills: leavers' lines, unused lines, roaming, premium charges, oversized plans

Sending a cancellation list

fairwire-ai-usage

AI usage: repeated prompts, large prompts, oversized models, cut-off answers, untagged spend

Sending an owner note

fairwire-api-retries

API and LLM retries: ignored Retry-After, retried client errors, retry storms, retry spend

Sending an owner note

Not here on purpose, because the Fairbooks agents already do them: certificates and domain expiry, mail authentication, MCP server review, configuration drift, access recertification and SaaS seats.

How they behave

  • Read-only on your systems. They read exported files under one folder (--root). They never log in to a firewall, controller, identity provider or cloud console.

  • Drafts, then one click. Each agent drafts a note or change request marked DRAFT. Approving it is one named person's click, tied to the draft's sha256, even in Auto mode; the person then makes the change. Live actions (pushing rules, disabling accounts, wiping devices, stopping cloud resources) are listed and refused.

  • Personal data masked. Names, sign-in names, e-mail addresses, phone numbers, device ids and client addresses become stable pseudonyms before results reach the model (--pii mask, the default). Files written for the team keep the real values.

  • Accepted as known. A finding a named person reviewed can be accepted with a reason and a review date; it then stops counting as open.

  • Files from other countries. Exports in nine languages read the same way: legacy encodings, 1.234,56, day-first dates and local column names. Report pages follow --lang de|fr|es|pt|it|ja|zh|ko.

  • Audit log. Every call is written to a hash-chained log under .fairwire/.

Checks on this repository

npm test                        # every workspace
npm run evals                   # every agent's eval cases
npm run hostile                 # planted formulas, HTML, path tricks and AI-directed text in every fixture
npm run hostile:args            # broken arguments in every tool
npm run intl                    # every spreadsheet eval case in eight more languages
npm run bench                   # 1,000 and 10,000 rows per agent
npm run release:check:offline   # metadata, honest-copy scan, secret scan, npm pack

Results and decisions are in REVIEW-ver1.md. Not legal, compliance or security-certification advice.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers