denpex-mcp 0.2.15 → 0.2.17

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +11 -11
  2. package/dist/index.js +2474 -2183
  3. package/package.json +55 -55
package/README.md CHANGED
@@ -1,12 +1,12 @@
1
1
  # Denpex MCP server
2
2
 
3
- Diagnose GPU/ML training failures — CUDA OOM, NCCL timeouts, Xid faults, NaN loss,
4
- checkpoint corruption — **without leaving your AI tool**. Ask Claude Desktop, Cursor,
3
+ Diagnose GPU/ML training failures. CUDA OOM, NCCL timeouts, Xid faults, NaN loss,
4
+ checkpoint corruption, **without leaving your AI tool**. Ask Claude Desktop, Cursor,
5
5
  or Claude Code to *"diagnose the crash in `./logs/train.err`"* and Denpex reads the log
6
6
  straight off disk and hands back the root cause and the fix.
7
7
 
8
8
  - 🩺 **Root cause + escalating fix** (primary → fallback → emergency), verification command, and prevention.
9
- - 🧭 **Find the culprit rank** in a distributed/NCCL hang, and route GPU Xid faults to the right node action — offline, instantly.
9
+ - 🧭 **Find the culprit rank** in a distributed/NCCL hang, and route GPU Xid faults to the right node action, offline, instantly.
10
10
  - 📚 **Bundled failure encyclopedia** (500+ documented classes) searchable by error code or symptom.
11
11
  - 🔒 **Zero-egress mode** (`DENPEX_LOCAL=1`): diagnose entirely on-host, nothing leaves the machine. Plus client-side PII scrubbing even in cloud mode.
12
12
 
@@ -75,7 +75,7 @@ claude mcp add denpex -e DENPEX_API_KEY=dpx_your_key_here -- npx -y denpex-mcp
75
75
 
76
76
  ### 3. Use it
77
77
 
78
- > "My training job crashed — diagnose the log at `./logs/rank0.err`."
78
+ > "My training job crashed, diagnose the log at `./logs/rank0.err`."
79
79
  > "What does Xid 79 mean?"
80
80
  > "This 64-GPU run is hung. Which rank failed first?" *(→ `analyze_cascade`)*
81
81
 
@@ -85,7 +85,7 @@ Or pick the **`diagnose-training-crash`** prompt your client surfaces.
85
85
 
86
86
  ## Offline / air-gapped mode
87
87
 
88
- Diagnose entirely on-host — no network, no logs leaving the cluster. Ideal for VPCs,
88
+ Diagnose entirely on-host, no network, no logs leaving the cluster. Ideal for VPCs,
89
89
  secure data centers, and privacy-sensitive labs. Add `DENPEX_LOCAL=1`:
90
90
 
91
91
  ```json
@@ -115,20 +115,20 @@ are unavailable offline.
115
115
  | `recent_diagnoses` | Recall your recent diagnoses and their fixes. | API key |
116
116
  | `ask_followup` | Conversational follow-up on a diagnosis. | Team plan / trial |
117
117
  | `check_versions` | Flag known-bad CUDA/PyTorch/NCCL/driver combos. | Team plan / trial |
118
- | `compare_runs` | "It worked yesterday" — isolate what changed between a good run and a bad one. | Team plan / trial |
118
+ | `compare_runs` | "It worked yesterday", isolate what changed between a good run and a bad one. | Team plan / trial |
119
119
  | `diff_environments` | Diff torch/CUDA/NCCL/driver versions and env vars between two runs. | Team plan / trial |
120
120
  | `validate_checkpoint` | Find the newest checkpoint that is actually safe to resume from. | Team plan / trial |
121
- | `preflight_check` | "Is this cluster safe to launch on?" — blockers before you burn GPU-hours. | Scale plan |
121
+ | `preflight_check` | "Is this cluster safe to launch on?", blockers before you burn GPU-hours. | Scale plan |
122
122
  | `node_health` | Score every monitored node, worst first. | Scale plan |
123
123
  | `predict_failures` | Nodes trending toward a failure that has not happened yet. | Scale plan |
124
124
  | `list_incidents` | Incidents Denpex's monitor caught & escalated on your fleet. | Scale plan |
125
125
 
126
- Gated tools degrade gracefully — they tell you exactly which plan unlocks them and which
126
+ Gated tools degrade gracefully, they tell you exactly which plan unlocks them and which
127
127
  tools work on your current plan. The failure encyclopedia is also exposed as resources at
128
128
  `denpex://failures/{slug}`.
129
129
 
130
130
  **Structured output.** Every tool above returns `structuredContent` alongside its
131
- markdown, validated against a declared output schema — so an agent branches on
131
+ markdown, validated against a declared output schema, so an agent branches on
132
132
  `confidenceBand === "low"` or iterates `rankedHypotheses` instead of regexing prose.
133
133
  Two fields deserve attention when you write the agent loop:
134
134
 
@@ -145,7 +145,7 @@ Two fields deserve attention when you write the agent loop:
145
145
 
146
146
  | Variable | Default | Purpose |
147
147
  |----------|---------|---------|
148
- | `DENPEX_API_KEY` | — | Your `dpx_…` key. Enables cloud tools. |
148
+ | `DENPEX_API_KEY` |, | Your `dpx_…` key. Enables cloud tools. |
149
149
  | `DENPEX_API_URL` | `https://api.denpex.com` | Override the API base (self-host / staging). |
150
150
  | `DENPEX_LOCAL` | `0` | `1` → diagnose entirely on-host (zero egress). |
151
151
  | `DENPEX_PRIVACY` | `off` | `mask` / `strict` → scrub PII from logs **before** they leave the machine. |
@@ -163,4 +163,4 @@ npm run inspector # interactive protocol inspector (MCP Inspector)
163
163
 
164
164
  The offline corpus (`src/data/denpex_patterns.json`) and canonical engine
165
165
  (`src/offline/canonical-engine.mjs`) are generated from the repo's single source of truth
166
- during `npm run build` — never hand-edit either artifact.
166
+ during `npm run build`, never hand-edit either artifact.