torjus/labmcp
Archived
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feature/lab-monitoring #6

Merged
torjus merged 7 commits from feature/lab-monitoring into master 2026-02-04 22:48:23 +00:00
Conversation 0 Commits 7 Files Changed 19 +2705 -23
torjus commented 2026-02-04 22:25:42 +00:00
Owner
Copy Link

Summary

New MCP server (lab-monitoring) for querying Prometheus and Alertmanager HTTP APIs. Unlike the existing nixpkgs/options servers, this one queries live APIs directly — no database or indexing required.

Tools (8 total)

Tool API Description
list_alerts Alertmanager List alerts with optional filters (state, severity, receiver)
get_alert Alertmanager Get full details for a specific alert by fingerprint
search_metrics Prometheus Search metric names with substring filter, enriched with metadata
get_metric_metadata Prometheus Get type, help text, and unit for a specific metric
query Prometheus Execute instant PromQL query
list_targets Prometheus List scrape targets with health status
list_silences Alertmanager List active/pending silences
create_silence Alertmanager Create a silence (gated behind --enable-silences flag)

MCP core changes (internal/mcp/server.go)

  • Added ModeCustom server mode for MCP servers without database dependencies
  • Added NewGenericServer(logger, config) constructor (no store parameter)
  • Added RegisterTool(tool, handler) method for external tool registration
  • Added InstructionsFunc callback on ServerConfig for dynamic per-session instructions

Dynamic alert instructions

On MCP initialization, the server calls AlertSummary() to check for active non-silenced alerts. If any are firing, the count and severity breakdown are included in the server instructions so the LLM can proactively inform the user.

Security: create_silence opt-in

The create_silence tool is a write operation that can suppress alerts. It is disabled by default and requires explicit opt-in:

  • CLI: lab-monitoring serve --enable-silences
  • NixOS module: services.lab-monitoring.enableSilences = true;

New files

  • internal/monitoring/ — Full package with types, Prometheus client, Alertmanager client, markdown formatters, MCP handlers
  • cmd/lab-monitoring/main.go — CLI with serve, alerts, query, targets, metrics commands
  • nix/lab-monitoring-module.nix — NixOS module (DynamicUser, no state directory needed)

Test coverage

  • prometheus_test.go — httptest-based tests for all Prometheus client methods
  • alertmanager_test.go — httptest-based tests for all Alertmanager client methods
  • handlers_test.go — End-to-end handler tests through the MCP server, including tool count verification with and without silences enabled

Other changes

  • Version bump: existing binaries 0.2.0 → 0.2.1 (shared internal/mcp/server.go change)
  • Added lab-monitoring package and lab-monitoring-mcp NixOS module to flake.nix
  • Updated .mcp.json with dev config entry
  • Updated CLAUDE.md and TODO.md

Stats

19 files changed, ~2700 lines added

## Summary New MCP server (`lab-monitoring`) for querying Prometheus and Alertmanager HTTP APIs. Unlike the existing nixpkgs/options servers, this one queries live APIs directly — no database or indexing required. ### Tools (8 total) | Tool | API | Description | |------|-----|-------------| | `list_alerts` | Alertmanager | List alerts with optional filters (state, severity, receiver) | | `get_alert` | Alertmanager | Get full details for a specific alert by fingerprint | | `search_metrics` | Prometheus | Search metric names with substring filter, enriched with metadata | | `get_metric_metadata` | Prometheus | Get type, help text, and unit for a specific metric | | `query` | Prometheus | Execute instant PromQL query | | `list_targets` | Prometheus | List scrape targets with health status | | `list_silences` | Alertmanager | List active/pending silences | | `create_silence` | Alertmanager | Create a silence (gated behind `--enable-silences` flag) | ### MCP core changes (`internal/mcp/server.go`) - Added `ModeCustom` server mode for MCP servers without database dependencies - Added `NewGenericServer(logger, config)` constructor (no store parameter) - Added `RegisterTool(tool, handler)` method for external tool registration - Added `InstructionsFunc` callback on `ServerConfig` for dynamic per-session instructions ### Dynamic alert instructions On MCP initialization, the server calls `AlertSummary()` to check for active non-silenced alerts. If any are firing, the count and severity breakdown are included in the server instructions so the LLM can proactively inform the user. ### Security: `create_silence` opt-in The `create_silence` tool is a write operation that can suppress alerts. It is disabled by default and requires explicit opt-in: - CLI: `lab-monitoring serve --enable-silences` - NixOS module: `services.lab-monitoring.enableSilences = true;` ### New files - `internal/monitoring/` — Full package with types, Prometheus client, Alertmanager client, markdown formatters, MCP handlers - `cmd/lab-monitoring/main.go` — CLI with `serve`, `alerts`, `query`, `targets`, `metrics` commands - `nix/lab-monitoring-module.nix` — NixOS module (DynamicUser, no state directory needed) ### Test coverage - `prometheus_test.go` — httptest-based tests for all Prometheus client methods - `alertmanager_test.go` — httptest-based tests for all Alertmanager client methods - `handlers_test.go` — End-to-end handler tests through the MCP server, including tool count verification with and without silences enabled ### Other changes - Version bump: existing binaries 0.2.0 → 0.2.1 (shared `internal/mcp/server.go` change) - Added `lab-monitoring` package and `lab-monitoring-mcp` NixOS module to `flake.nix` - Updated `.mcp.json` with dev config entry - Updated `CLAUDE.md` and `TODO.md` ### Stats 19 files changed, ~2700 lines added
torjus added 3 commits 2026-02-04 22:25:42 +00:00
feat: add lab-monitoring MCP server for Prometheus and Alertmanager 1755364bba 
New MCP server that queries live Prometheus and Alertmanager HTTP APIs
with 8 tools: list_alerts, get_alert, search_metrics, get_metric_metadata,
query (PromQL), list_targets, list_silences, and create_silence.

Extends the MCP core with ModeCustom and NewGenericServer for servers
that don't require a database. Includes CLI with direct commands
(alerts, query, targets, metrics), NixOS module, and comprehensive
httptest-based tests.

Bumps existing binaries to 0.2.1 due to shared internal/mcp change.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
feat: include active alert count in MCP server instructions 2a08cdaf2e 
Add InstructionsFunc callback to ServerConfig, called during each
initialize handshake to generate dynamic instructions. The lab-monitoring
server uses this to query Alertmanager and include a count of active
non-silenced alerts, so the LLM can proactively inform the user.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
feat: gate create_silence behind --enable-silences flag 06e62eb6ad 
The create_silence tool is a write operation that can suppress alerts.
Disable it by default and require explicit opt-in via --enable-silences
CLI flag (or enableSilences NixOS option) as a safety measure.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
torjus added 2 commits 2026-02-04 22:33:53 +00:00
docs: add list_rules and get_rule_group to lab-monitoring TODO 741f02d856 
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
chore: update monitoring URLs to production endpoints 5b9eda48f8 
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
torjus added 2 commits 2026-02-04 22:46:41 +00:00
docs: add Loki log query support to lab-monitoring TODO d31a93d3b6 
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
chore: enable silences in lab-monitoring MCP config 52f50a1a06 
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
torjus merged commit b491a60105 into master 2026-02-04 22:48:23 +00:00
torjus deleted branch feature/lab-monitoring 2026-02-04 22:48:24 +00:00
This repo is archived. You cannot comment on pull requests.
Reviewers
No Reviewers
Labels
No items
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
torjus
No Assignees
1 Participants
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: torjus/labmcp#6
Delete Branch "feature/lab-monitoring"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?