simplify MCPHub extensions and restore Hermes web tools

This commit is contained in:
Mikei386
2026-08-26 11:08:47 +02:00
parent 9455a79dc2
commit bc902261b9
14 changed files with 503 additions and 226 deletions
+6 -3
View File
@@ -36,8 +36,11 @@ terminal:
lifetime_seconds: 1800
web:
search_backend: "searxng"
extract_backend: "native"
# Keenable is bundled with Hermes and provides both search and extraction
# through its keyless tier. This remains available after Hermes moved to
# Unraid, where the former Athena-local SearXNG no longer exists.
search_backend: "keenable"
extract_backend: "keenable"
extract_char_limit: 15000
keyless_fallback: true
keyless_rescue: true
@@ -161,7 +164,7 @@ skills:
creation_nudge_interval: 20
plugins:
enabled: ["web-searxng"]
enabled: ["browser-browser-use", "web-keenable"]
timeouts:
tools:
-1
View File
@@ -42,7 +42,6 @@ ROUTER_API_KEY=$router_key
VOICE_TOOLS_OPENAI_KEY=$router_key
MUA_MCP_URL=$mua_url
MUA_MCP_BEARER_TOKEN=$mua_token
SEARXNG_URL=http://searxng:8080
API_SERVER_ENABLED=true
API_SERVER_HOST=0.0.0.0
API_SERVER_PORT=8642
+11 -2
View File
@@ -34,6 +34,12 @@ create_profile() {
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set agent.reasoning_effort low
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set display.show_reasoning false
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set auxiliary.title_generation.enabled false
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set web.search_backend keenable
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set web.extract_backend keenable
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set web.keyless_fallback true
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set web.keyless_rescue true
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set plugins.enabled \
'["browser-browser-use","web-keenable"]'
docker exec "$HERMES_CONTAINER" hermes -p "$name" config unset compression.threshold_tokens || true
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set compression.threshold 0.95
docker exec "$HERMES_CONTAINER" hermes -p "$name" config set compression.target_ratio 0.15
@@ -72,7 +78,9 @@ done < <(jq -r --argjson max "$(jq '.max_output_tokens' "$PROFILE_MATRIX")" \
# The Unraid host deliberately has no system Python. Run the small declarative
# client renderer in the already version-pinned MCPHub image instead of adding
# host dependencies.
sync_args=(--registry /stack/config/mcp-registry.json)
mcphub_registry=${MCPHUB_REGISTRY:-/mnt/nvme-storage/appdata/MCPHub/config/mcp-registry.json}
[[ -s $mcphub_registry ]] || die "MCPHub-Registry fehlt: $mcphub_registry"
sync_args=(--registry /run/input/mcp-registry.json)
mcphub_token=${MCPHUB_TOKEN_FILE:-/mnt/nvme-storage/appdata/MCPHub/client-token}
token_mount=()
if [[ -s $mcphub_token ]]; then
@@ -85,8 +93,9 @@ done < <(find "$HERMES_DATA_DIR" -name config.yaml -type f -print)
docker run --rm --entrypoint python \
-v "$STACK_DIR:/stack:ro" \
-v "$HERMES_DATA_DIR:/hermes:rw" \
-v "$mcphub_registry:/run/input/mcp-registry.json:ro" \
"${token_mount[@]}" \
casaderoll/mcphub:1.1.0 \
casaderoll/mcphub:1.2.1 \
/stack/platform/mcp/sync-clients.py "${sync_args[@]}"
"$STACK_DIR/platform/hermes/install-skills.sh"
+86 -208
View File
@@ -1,229 +1,107 @@
---
name: mcphub-deployer
description: Install, update, disable, test, publish, or remove portable MCP servers in CasaDeRoll MCPHub on Unraid. Use for MCP catalog, GitHub, npm, PyPI, Go-binary, existing HTTP-MCP, client-registration, MCPHub repair, or moving an MCP out of Athena or Hermes.
description: Install, update, disable, test, publish, or inspect portable MCP servers in CasaDeRoll MCPHub on Unraid. Use for MCP repositories, packages, binaries, HTTP MCPs, client registration, MCPHub repair, or moving an MCP out of Athena or Hermes.
---
# MCPHub Deployer
Install portable MCPs in the existing `MCPHub` container on Unraid. Never
create one Docker container per portable MCP and never install one inside
Hermes. Hermes, OpenWebUI, Pi, and other agents are clients of MCPHub.
Install portable MCPs in the existing `MCPHub`; never create a separate
container and never install them inside Hermes.
## Fixed production map
## Fixed map
Use these paths directly. Do not search the filesystem for alternatives.
- Unraid: `192.168.1.2`; container: `MCPHub`; base URL: `http://192.168.1.2:8787`
- Appdata: `/mnt/nvme-storage/appdata/MCPHub`
- Work: `/mnt/nvme-storage/appdata/MCPHub/work/<id>`
- Extensions: `/mnt/nvme-storage/appdata/MCPHub/extensions/<id>`
- Registry: `/mnt/nvme-storage/appdata/MCPHub/config/mcp-registry.json`
- Secret: `/mnt/nvme-storage/appdata/MCPHub/secrets/<id>.env` (0600)
- Helper in container: `/opt/casaderoll/deploy-extension.py`
- Route: `http://192.168.1.2:8787/mcp/<id>`
- Host: Unraid `192.168.1.2`
- Container: `MCPHub`
- UI/base URL: `http://192.168.1.2:8787`
- Operational source/build tree: `/mnt/nvme-storage/appdata/MCPHub/build/repo`
- Dockerfile: `/mnt/nvme-storage/appdata/MCPHub/build/repo/platform/mcphub/Dockerfile`
- Single server and client registry: `/mnt/nvme-storage/appdata/MCPHub/build/repo/config/mcp-registry.json`
- Registry renderer: `/mnt/nvme-storage/appdata/MCPHub/build/repo/platform/mcphub/configure-settings.py` (normally unchanged)
- Versioned Unraid-template source: `/mnt/nvme-storage/appdata/MCPHub/build/repo/config/unraid-templates/my-MCPHub.xml`
- Live DockerMan template: `/boot/config/plugins/dockerMan/templates-user/my-MCPHub.xml`
- Persistent state: `/mnt/nvme-storage/appdata/MCPHub`
- Secrets: `/mnt/nvme-storage/appdata/MCPHub/secrets/<server>.env`, mode `0600`
- Client bearer token: `/mnt/nvme-storage/appdata/MCPHub/client-token`
- Individual route: `http://192.168.1.2:8787/mcp/<server>`
Do not search for other checkouts, registries, templates, or secret stores.
Normal extensions do not modify Dockerfile, image tag, Unraid template, MCPHub
source, or Hermes config manually.
The versioned template and the live DockerMan template are different files.
Keep their image tag and required mounts aligned. If the versioned template is
missing, copy the live template to that exact versioned path once; do not search
for another template and do not infer a replacement from unrelated containers.
## Hard credential boundary
The operational build tree is persistent and covered by the normal Unraid
Appdata backup. The checkout below is legacy and MUST NOT be used or inspected
for MCPHub work:
Credentials are user input. Check only whether the dedicated secret file and
required keys exist; never print values. Never inspect other containers,
environments, mail servers, configs, histories, or passwords to find or infer
credentials. If credentials are missing, complete credential-independent build
work, stage the MCP disabled, report the exact secret path and missing key
names, then stop. Never publish or authenticate it.
`/mnt/nvme-storage/Eigene Dateien/Michael/Entwicklung/AI-Profile-Router`
## One-pass workflow
The normal Git repository remains the durable documentation/history. Publish
the same focused files there when an authorized Git write path is available.
Lack of Git access is a warning to report, not permission to search for other
checkouts and not a reason to abandon an otherwise requested local install.
1. Read this skill once. Inspect only MCPHub state, the fixed registry, the
dedicated secret-file presence, and the target upstream release/source.
2. Classify once: existing HTTP MCP, packaged stdio MCP, released binary,
custom MCP, or host-bound HTTP proxy. Do not reconsider without a concrete
failed build or handshake.
3. Interpret intent: “prüfen/planen” changes nothing; “installieren/einbauen”
continues; “deaktiviert” never activates; “read-only” omits mutating tools.
4. Build only in `/tmp` or the fixed work directory. Pin versions and verify
checksums. Put runtime files in the work directory, not in the repository.
5. Write one compact manifest at `<work>/manifest.json`:
## Mandatory fast path
```json
{
"server": {
"id": "example", "hermes_id": "example", "name": "Example",
"description": "Short tool-selection description",
"url": "http://192.168.1.2:8787/mcp/example",
"clients": ["hermes"],
"deployment": {"required_env": ["EXAMPLE_TOKEN"]},
"hub": {
"type": "stdio", "secret_file": "example.env",
"command": "/usr/local/bin/run-with-env",
"args": ["/run/secrets/mcphub/example.env", "--", "node", "/app/data/extensions/example/index.js"],
"enabled": false
}
},
"artifacts": [
{"source": "/app/data/work/example/index.js", "path": "index.js", "sha256": "HEX", "mode": "0644"}
]
}
```
For a normal installation, perform these phases once and in order. Do not
restart discovery after a phase has completed.
Use paths as seen inside MCPHub (`/app/data/...`) in the manifest.
6. Stage with exactly:
### 1. Preflight — at most six checks
```sh
docker exec MCPHub python3 /opt/casaderoll/deploy-extension.py stage \
--manifest /app/data/work/<id>/manifest.json
```
Check only:
The helper copies verified files, updates the registry atomically, and always
stages disabled. It technically blocks activation when the dedicated secret
or a required key is missing.
7. Recreate only `MCPHub` once so it reconciles the external registry. Do not
restart Hermes, Athena, Router, Qwen, WireGuard, Unraid, or other services.
8. If credentials are ready, activate with the helper, recreate only MCPHub,
then verify: health; all old routes; new handshake; `list_tools` schemas;
one bounded read-only call; no test writes or residue. Otherwise stop while
disabled.
9. Run the existing client-sync script only after successful activation. Do
not edit client YAML by hand. New routes are not advertised automatically.
10. Report version, state, tool count, secret path (never values), tests,
client sync, durable source status, and rollback.
1. `MCPHub` container state, image tag, mounts, and network.
2. The fixed Dockerfile, registry, renderer, and both template paths above.
3. Existing MCPHub server names to avoid duplication.
4. Target service reachability or the upstream release.
5. Required secret-file presence; never print its values.
6. Current Git availability, if any.
## Limits
Never run `find` to rediscover a listed path. Never read unrelated Compose
stacks, repositories, documentation trees, container logs, container
environments, application configs, home directories, or secret stores.
- At most six preflight reads and two attempts per hypothesis.
- At most one corrected build.
- Never dump full registries, repository trees, logs, or configs; use bounded
queries and compact JSON.
- Never use one giant shell call to write several files.
- Never claim success without observed handshake and probe results.
- On failure leave the previous MCPHub running and the new extension disabled.
- Ask before destructive actions or credential rotation not explicitly asked.
### Credential boundary — mandatory stop
## Resume
Credentials and account configuration are user-supplied inputs, not discovery
targets. This rule overrides the desire to complete a deployment or live test.
- Never inspect another container, service, environment, mount, config file,
mailbox, shell history, password manager, or secret directory to obtain or
infer credentials.
- Never reuse credentials found in an existing mail server or another app
unless the user explicitly names that exact source and authorizes reuse.
- Check only whether the dedicated fixed secret file for this server exists;
do not read its values during preflight.
- If the MCP needs credentials or account settings and the dedicated file is
absent or incomplete, finish all credential-independent build work, install
the server disabled, and stop before live authentication. Ask the user for
the missing fields and state the exact secret-file path.
- Do not substitute inspection of an existing service for that question.
- A missing credential may reduce verification to build plus MCP handshake; it
is never permission to investigate the user's infrastructure.
### 2. Classify once
Choose exactly one integration:
- Existing HTTP MCP: declare its URL and authentication; do not copy it.
- Packaged stdio MCP: pin and install the exact npm/Python package in MCPHub.
- Released binary: pin version and checksum; download and verify it during the
Docker image build. Do not commit a downloaded binary.
- Custom MCP: keep source in the build tree and copy it into the image.
- Host-bound MCP: leave it on its required host and proxy its authenticated
HTTP endpoint through MCPHub.
Do not reconsider this classification unless a real build or handshake result
contradicts it.
### 3. Interpret the user's intent
- “Prüfe/plane/zeige den Ablauf”: inspect and return a short plan; change
nothing.
- “Installiere/baue ein/los/Abfahrt”: continue through deployment and tests.
- “Zunächst deaktiviert”: install the runtime and declaration with
`enabled: false`; do not add it to client registries yet.
- “Nur lesen”: disable or omit mutating tools before client publication.
Ask only for information that cannot be derived safely: credentials, a
material license decision, or an ambiguous destructive permission.
For mail-related MCPs, repository inspection may determine which field names
are required, but the IMAP/SMTP host, user, password/token, sender identity, and
TLS choices must come from the user or the dedicated secret file. The presence
of a Docker mail server does not answer those questions.
### 4. Implement the smallest change
Modify only the necessary fixed production files. Rules:
- Pin image, package, release, and checksum versions.
- Put credentials only in the matching secret file with mode `0600`.
- Never print, log, commit, summarize, or return a secret.
- Use `/usr/local/bin/run-with-env` for secret-backed stdio servers.
- Declare servers only in `config/mcp-registry.json`; do not hard-code a server
in `configure-settings.py` and do not edit `mcp_settings.json` manually.
- Preserve existing users, bearer keys, prompts, resources, enabled states,
and per-tool toggles.
- Build a new image tag. Never overwrite the tag currently running.
- Update both the versioned template source and the live DockerMan template to
the exact new tag. Do not rebuild `template.xml` inside an upstream image.
For servers exposing many tools, install disabled first. After a successful
local test, enable only the required tool groups in MCPHub. Do not publish an
unfiltered large server to clients.
### 5. Deploy without collateral changes
Build from `/mnt/nvme-storage/appdata/MCPHub/build/repo`, then recreate only
`MCPHub` through Unraid DockerMan so it stays a managed Unraid container.
Preserve all Appdata and mounts.
Never restart Athena, Router, Qwen, Hermes, OpenWebUI, WireGuard, Unraid, or
unrelated containers for an MCPHub deployment.
Do not use unrestricted host shell access during repository analysis,
classification, or credential preflight. Use it only after the exact file
changes, new image tag, verification steps, and rollback tag are known. Its
scope is then limited to those declared paths and the `MCPHub` container.
### 6. Prove the result
Verify, in this order:
1. New container uses the intended image and remains healthy.
2. Existing MCP routes still handshake.
3. New server starts when enabled.
4. MCP handshake and `list_tools` succeed with valid schemas.
5. One bounded read-only function returns plausible live data.
6. No test download, queue item, write, or second backend remains.
7. If installed disabled, return it to disabled after the temporary test.
If credentials are unavailable, steps 3 and 5 may be recorded as blocked.
Never weaken the credential boundary merely to make the live probe pass.
A running container alone is not success. Never claim install, test,
registration, Git push, or backup without observing its result.
### 7. Publish to clients only after filtering
Add `http://192.168.1.2:8787/mcp/<server>` to
`config/mcp-registry.json` only after the server and selected tools have passed
verification. Generate intended Hermes/OpenWebUI registrations from that one
registry. Reload only the affected client gateway if required.
New MCPHub servers are not automatically advertised by the model router.
### 8. Finish compactly
Report exactly:
- installed version and image tag;
- enabled/disabled state and exposed tool count;
- secret-file path without values;
- handshake and read-only probe result;
- client registrations changed or intentionally omitted;
- Git status/push result;
- rollback image tag.
Do not narrate repeated planning or internal reconsideration.
## Context-compaction checkpoint
Before a long build or whenever context use approaches compression, write a
small checkpoint to:
`/mnt/nvme-storage/appdata/MCPHub/work/<server>.json`
Store only: requested outcome, integration type, completed phases, modified
paths, old/new image tags, pending action, verification results, and rollback.
Never store credentials. After compression, reread this SKILL.md directly from
`/opt/data/skills/platform/mcphub-deployer/SKILL.md`
inside the Hermes container plus that checkpoint, then continue at the pending
phase. Do not rely on a deduplicated/pruned earlier skill result and never
rediscover completed phases.
Delete the checkpoint after a successful final report. Keep it on failure so a
new session can resume safely.
## Hard limits and stop rules
- Maximum two attempts for the same command, endpoint, or hypothesis.
- Maximum one corrected build after the first failed build.
- Never repeat “I now understand the mechanism” and continue researching.
- If upstream command, transport, license, or credentials remain unknown after
two focused checks, stop and state that exact blocker.
- If deployment fails, keep the previous container/image running and report the
failing phase. Do not improvise a second container or unversioned binary.
- Ask before destructive queue actions, downloads, service mutations, or
credential rotation that the user did not explicitly authorize.
## Rollback
Restore the previous image tag and declarative server entry, recreate only
`MCPHub`, and repeat existing-route handshakes plus one read-only probe. Never
delete Appdata or shared secret files during rollback.
Before compression write `<work>/checkpoint.json` containing only phase,
version, completed checks, pending action, modified paths, and rollback. After
compression reload this skill and that checkpoint before continuing. Delete
the checkpoint only after success.