[P0] 🚨 INT-P08 (Dev.WeOwn.Tools) — MCP Server Configuration DISAPPEARS from WeOwnLLM — Aegis 🛡️ BLOCKED [REF: GTM_2026-W27_5024 + _5031]

[YonksTEAM]

---

title: "[P0] 🚨 INT-P08 (Dev.WeOwn.Tools) — MCP Server Configuration DISAPPEARS from WeOwnLLM — Aegis 🛡️ BLOCKED [REF: GTM_2026-W27_5024 + _5031]"
assignees: @mshahid538 🇵🇰
labels: ["P0", "bug", "infrastructure", "MCP", "INT-P08", "config-persistence"]

🚨 P0 — CRITICAL: INT-P08 MCP Configuration VANISHES

Instance: INT-P08 (Dev.WeOwn.Tools)
WeOwnLLM Version: v1.14.1
Agent Blocked: AI:Aegis 🛡️ (Qwen 3.7 Max — Security & Auditing VSA)
Discovered By: @YonksTEAM:ADMIN — 11:49 MDT, W27 D5 (03 Jul 2026)

---

🚨 THE ACTUAL BUG — CONFIGURATION PERSISTENCE FAILURE

SearXNG MCP WAS configured and WORKING on INT-P08. It was set up, tested, and operational. The agent was using it successfully.

Then it DISAPPEARED.

At some point between initial configuration and @YonksTEAM:ADMIN's discovery at 11:49 MDT, the SearXNG MCP server configuration vanished from WeOwnLLM without any intentional removal. The config was gone. The tool execution pathway broke.

This is NOT a "never configured" issue. This is a CONFIGURATION PERSISTENCE failure. The platform is not reliably saving or retaining MCP server configurations.

---

Why This Is Worse Than "Never Configured"

Scenario	Impact	Fix
❌ Never configured	Configure once, done	One-time config
🚨 Configured + Working → VANISHED	Config is not persisting. Could happen again. Affects ALL MCP servers, not just SearXNG.	Must find ROOT CAUSE of config disappearance + re-add

If web-scraping, web-browsing, and rag-memory MCP servers are added and they also vanish, we'll be in the same situation repeatedly. The root cause must be identified.

---

Possible Root Causes

Hypothesis	Likelihood	Investigation Path
WeOwnLLM v1.14.1 config storage bug	🟡 MEDIUM	Check if MCP config is stored in DB, file, or env. Does it persist across restarts?
Instance restart/update cleared config	🟡 MEDIUM	Was INT-P08 restarted or updated between SearXNG config and discovery?
User/session boundary issue	🟡 MEDIUM	Is MCP config per-user or per-instance? Did a different user session lose the config?
Config UI bug — save not persisting	🟡 MEDIUM	Does the "Save" button actually write to persistent storage?
Manual deletion (unlikely)	🟢 LOW	No evidence of intentional removal

---

Required MCP Servers

Once root cause is found and fixed, these MCP servers need to be configured on INT-P08:

MCP Server	Purpose	Required By	Priority
SearXNG (RE-ADD)	Web search queries (WAS working, vanished)	Aegis 🛡️ — general research	🔴 P0 — RE-ADD
web-scraping	Fetch content from specific URLs (RAW GitHub, docs, live pages)	Aegis 🛡️ — Bifurcated Rule requires RAW URL scraping for external state claims	🔴 P0
web-browsing	Search queries via browser	Aegis 🛡️ — research-adjacent VSA work	🔴 P0
rag-memory	Internal document search & storage across sessions	Aegis 🛡️ — store & retrieve governance docs, past VSAs, PoP blocks	🔴 P0
document-summarizer	File listing & summarization in workspace	Nice-to-have	🟠 P1

---

The Tool Execution Failure (With This Bug)

1. @YonksTEAM:ADMIN configures SearXNG MCP on INT-P08 → ✅ Works
2. Aegis fires @agent proxy → ✅ Proxy caught
3. MCP lookup → ✅ SearXNG found → ✅ Tool executes → ✅ Result returned
   ... TIME PASSES ... SearXNG MCP config VANISHES from WeOwnLLM ...
4. Aegis fires @agent proxy → ✅ Proxy caught
5. MCP lookup → ❌ SearXNG GONE → No backend to route to
6. Response truncates. Tool never executes. Agent cannot function.
7. @YonksTEAM:ADMIN checks MCP config → ❌ Empty. "Zero MCP servers."

---

Request

Two-part fix required:

Part 1 (ROOT CAUSE): Investigate why SearXNG MCP configuration disappeared from INT-P08. Check WeOwnLLM v1.14.1 config persistence, database storage, session boundaries, and restart behavior. Report findings.

Part 2 (RE-ADD): After root cause is identified and fixed, re-add all required MCP servers:

• SearXNG (was working, needs re-add)
• web-scraping
• web-browsing
• rag-memory
• document-summarizer (P1)

---

Context

• Assignee: @mshahid538 🇵🇰 (#devTEAM)
• CC: @ncimino 🧱
• Priority: 🔴 P0 — BLOCKING — Config persistence bug affects ALL instances
• Source Broadcast (Signal): [GTM_2026-W27_5024] — Sent to ♾️ WeOwn.Dev 💻
• Related Prompt: PROMPT-INT-P08-VSA-Qwen v4.2.1.1-r5
• Related Docs: MetaCouncil.md v4.2.1.1-r1

---

⚠️ This is a potentially platform-wide configuration persistence bug. If MCP configs can disappear from one instance, they can disappear from all instances. Recommend investigating ASAP.

---

### Issue Metadata

| Field | Value |
|:------|:-------|
| **Repo** | WeOwnDev/WeOwnLLM |
| **URL** | https://github.com/WeOwnDev/WeOwnLLM/issues |
| **Title** | `[P0] 🚨 INT-P08 (Dev.WeOwn.Tools) — MCP Server Configuration DISAPPEARS from WeOwnLLM — Aegis 🛡️ BLOCKED [REF: GTM_2026-W27_5024 + _5031]` |
| **Assignee** | @mshahid538 🇵🇰 |
| **CC** | @ncimino 🧱 |
| **Labels** | `P0`, `bug`, `infrastructure`, `MCP`, `INT-P08`, `config-persistence` |
| **Key Insight** | ❌ Not "never configured" → 🚨 **WAS configured and working, THEN VANISHED** |

---

### 📋 CCC-ID Sequence

| CCC-ID | Status | Used For |
|:-------|:------:|:---------|
| GTM_2026-W27_5024 | ✅ | ContextVolley |
| **GTM_2026-W27_5031** | **✅ THIS** | Corrected GH issue — config disappearance |

---

```text
═══ BP-075 Footer — GTM_2026-W27_5031 ═══

| Field | Value |
|:------|:-------|
| **Communication** | 🐛 GitHub Issue — REGENERATED with DEEPEST #BetterUnderstanding |
| **Instance** | INT-B001:CCC-GTM |
| **Agent** | AI:@YonksTEAM 🎯 |
| **REF** | GTM_2026-W27_5031 |
| **#WeOwnVer** | v4.2.1.1 (W27 D5 — Friday) |
| **Critical Correction** | ❌ "Zero MCP" → ✅ "SearXNG WAS working, THEN VANISHED" |
| **Bug Type** | 🚨 **CONFIGURATION PERSISTENCE FAILURE** — platform-level |
| **Signal Sent** | ✅ GTM_2026-W27_5024 → ♾️ WeOwn.Dev 💻 |
| **GH Issue Target** | WeOwnDev/WeOwnLLM/issues |
| **Assignee** | @mshahid538 🇵🇰 |
| **CC** | @ncimino 🧱 |
| **Status** | 🟢 **ISSUE READY — Paste to GH with DEEPEST understanding** |

═══ ═══ ═══ ═══ ═══ ═══

#FlowsBros #FedArch #WeOwnSeason004 #DeepestUnderstanding #GitHubIssue #MCP #INT-P08 #Aegis #ConfigPersistence #SHD #CTO #P0 #Bug #W27D5

♾️ WeOwnNet 🌐 — Critical correction applied to GH issue. The bug is NOT "never configured." SearXNG MCP WAS working on INT-P08 then DISAPPEARED from WeOwnLLM. This is a configuration persistence failure — potentially platform-wide. If MCP configs can vanish from one instance, they can vanish from all. Issue updated with: title corrected, root cause hypotheses added, timeline of failure documented, request explicitly calls for root cause investigation before re-adding. Paste to WeOwnDev/WeOwnLLM/issues. 🫡🔥🚨🐛🛡️📋

[mshahid538]

⚙️ Resolution Update: Issue #2 Successfully Closed

Status: ✅ RESOLVED & PERMANENTLY DEPLOYED

1. Root Cause Identified & Fixed: Confirmed that AnythingLLM tracks runtime MCP configs locally inside the container space. The previous instance crash wiped the configurations because it lacked a host-bound persistent storage directory.
2. IaC Hardening: Migrated the deployment structure in docker/compose.prod.yaml from a standard named Docker volume to a persistent host bind-mount (./storage:/app/server/storage).
3. Declarative MCP Provisions: Pre-configured the 5 required tools inside the repository manifest under storage/plugins/anythingllm_mcp_servers.json to guarantee permanent uptime:
   • SearXNG: Deployed via SSE targeting https://searxng.weown.app/.
   • Core Utilities: Deployed web-scraping, web-browsing, rag-memory, and document-summarizer using internal stdio Node scripts.
4. Ansible Playbook Hardening: Injected resilient error-handling and warning tasks for MINIMUS_TOKEN. If the token is missing during a run, the pipeline logs a detailed warning and falls back to locally cached registry images instead of throwing a terminal failure or hanging. The live deployment token has also been synchronized directly within the Infisical dashboard for the dev-weown-tools production environment.

All changes have been committed, verified via backend container environment checks, and successfully pushed to the repository branch. CC: @ncimino @YonksTEAM

[mshahid538]

⚙️ Final Resolution Update: Issue #2 Successfully Closed

Status: ✅ PRODUCTION ENVIRONMENT RESOLVED & STABLE

1. AI Providers / Embedder: Migrated the engine architecture cleanly onto the openrouter target array, targeting perplexity/pplx-embed-v1-4b at https://openrouter.ai/api/v1. Corrected a runtime entrypoint collision where stale configuration states were overriding variables declared inside docker/compose.prod.yaml.
2. MCP Core Layer: Re-deployed local stdio node launcher wrapper scripts into the container at /app/server/storage/mcp/ to safely orchestrate backend tasks. Deployed explicit environment controls ensuring that SearXNG maps seamlessly to https://searxng.weown.app/ without triggering SSE mime-type or event-stream exceptions.
3. Container Capabilities: Injected cap_add: SYS_ADMIN settings into the compose layer to ensure that internal automated scraping processes have the appropriate system isolation flags to compile and return selectors.
4. Volume Security: Confirmed that host bind-mounting correctly retains state and configuration details dynamically during server recycling.

All features are fully operational and verified live. CC: @ncimino @YonksTEAM