Reference¶

`health_check()`¶

Seven keys are present on every call:

Key	Meaning
`status`	`healthy` / `degraded` / `error`
`service`	Always `eos-mcp`
`version`	Package version
`config_path`	Resolved `config.ini` path (present even when the load failed, so callers can see where it looked)
`device_count`	Number of `[hostname]` sections found
`tags`	Sorted, deduplicated `tags =` values across all devices
`config`	`ok` / `error` / `missing`

detail is added only on degraded or error, carrying the reason (a missing file, or a config.ini parse error).

Lightweight by design: it only reads and parses config.ini — it does not open any eAPI/pyeapi connection to an EOS device, so it is safe to call against a large fleet without touching the network. Every other tool needs a real, populated config.ini at the resolved path to do anything.

Tool index¶

Tool	Purpose
`get_router_list(tags=None)`	List devices registered in `config.ini`, optionally filtered by tag
`get_device_facts(hostname)`	Model, serial, EOS version, hardware revision, uptime, memory, MAC, architecture
`get_device_facts_batch(hostnames=None, tags=None, max_workers=5)`	Same facts for multiple devices in parallel
`get_version(hostname)`	Model + EOS version (quick connectivity check)
`run_command(hostname, command)`	Unrestricted. Run one enable-mode command on one device
`run_commands(hostname, commands)`	Unrestricted. Run multiple enable-mode commands on one device
`run_command_batch(command, hostnames=None, tags=None, max_workers=5)`	Unrestricted. Run one command across multiple devices in parallel
`run_commands_batch(commands, hostnames=None, tags=None, max_workers=5)`	Unrestricted. Run multiple commands across multiple devices in parallel
`get_config(hostname)`	Retrieve running-config
`get_config_diff(hostname, rollback_id=1)`	Diff running-config vs. the Nth rollback checkpoint
`list_config_sessions(hostname)`	List configure sessions and their state (pending / pendingCommitTimer / completed)
`push_config(hostname, config_lines, session_name="mcp-push", dry_run=True, commit_timer=300)`	Writes. Push config via a named configure session; dry-run by default
`confirm_config_session(hostname, session_name="mcp-push")`	Writes. Finalize a pending commit-timer session
`abort_config_session(hostname, session_name="mcp-push")`	Writes. Discard a pending session
`collect_tech_support(hostname)`	Collect `show tech-support` output (large, 30+ seconds)
`daily_brief(hostnames=None, tags=None, max_workers=5, since_hours=24)`	Morning fleet health check: environment, errdisabled interfaces, uptime, MLAG, recent syslog alerts

"Unrestricted" tools accept and execute any enable-mode command text verbatim — see Setup → Write operations for what that means in practice and what gates it.

`daily_brief`¶

Runs check_health() against every resolved device in parallel and renders one Markdown report: per-device CRITICAL / WARNING / OK status (environment sensors, errdisabled interfaces, MLAG state, memory, and syslog alerts matching BGP/OSPF/STP/LACP/MLAG/link-down patterns within the last since_hours hours), followed by a fleet-wide summary count. A connection failure to a device renders as a CRITICAL: connection failed line for that device rather than dropping it from the report silently.

Targets are resolved from hostnames, tags, or both; if neither is given, every device in config.ini is used.

`get_config_diff`¶

rollback_id=1 (the default) diffs against the most recent rollback checkpoint; a higher rollback_id reaches further back. Diffing against startup-config requires EOS 4.30+ — older releases fall back gracefully.

CLI¶

eos-mcp                              # start the MCP server (stdio; default, no arguments)
eos-mcp -V | --version                # print version and exit
eos-mcp -h | --help                   # show usage
eos-mcp --check                       # verify config.ini and list devices, then exit
eos-mcp --check --check-host HOST     # also open an eAPI connection to HOST

Exit codes for --check / --check-host: 0 success, 1 a configuration error (config.ini missing or unparsable), 2 a host-connection error (host not found in config, or the eAPI call itself failed).

TLS compatibility¶

EOS 4.28.x combined with Python's stricter default TLS policy (notably under Python 3.14) can raise SSLV3_ALERT_HANDSHAKE_FAILURE. eos-mcp patches the SSL context at import time (SECLEVEL=0, minimum TLS 1.0) so older devices stay reachable without operator intervention. verify = false is the default in config.ini.example for the same reason — most fleets in the wild run internal, self-signed eAPI certificates.