Model Manager

Manage your local model machine over MCP: state, model pull/remove, service switch, LoRA training

Connector by setheerwagen · GitHub account verified by the MCP Registry

IndexedMITLocal (stdio)

Chalk Labs

Need help deploying Model Manager?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Codex, VS Code, Cursor, Windsurf, Claude Desktop. Read the repository before you give any tool access to your data.

npx agentmarket install connector/setheerwagen-mcp-modelmanager

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

About

Manage your local model machine over MCP: state, model pull/remove, service switch, LoRA training

Reviews

Write a review

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this server can do and what it asks for, from its tool list, install config and code. Checked Oct 25, 2026.

Tools: 10 read · 5 write · 2 delete · 2 not yet classified
Secrets: None asked for
Install scripts: None found
Runs as: Python package

Tools by what they do 19

Read from the source code; the running server may list different tools.

Read 10 · Look things up without changing anything.

check_isolation
check_training_env
fetch_model
job_status
list_adapters
list_jobs
list_variants
show_variant
status
training_status

Write 5 · Create, change, send or run things.

cancel_training
create_variant
start_service
start_training
stop_service

Delete 2 · Delete, drop, revoke or wipe things.

remove_model
remove_variant

Not yet classified 2 · We could not tell from the name and description.

models_present
switch_service

Secrets and settings it asks for

Other settings (environment variables or headers).

MM_CONTAINER_ROOT
MM_ENABLED_OPERATIONS
MM_MAX_TIER
MM_VM2_HOST
MM_VM_HOST
MM_VM_USER

Network hosts it talks to 1

github.com · found in code

Files it touches

~/.ssh/modelmanager
/dev/null
/dev/x
/etc/shadow
/mnt/nvme/hf
/root/.ollama
/root/.ollama/adapter/run1
/srv/models
/srv/models/../etc/passwd
/srv/models/adapter/run1
/srv/models/adapter/try1
/srv/models/data
/srv/models/data/train.jsonl
/srv/models/ollama/adapter/../../etc
/srv/models/ollama/adapter/run1
/srv/modelsX/data
Your home directory (~)
~/.local/state
Any path the agent passes to its tools

Tools (19)

Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].

cancel_training
Cancels a running training run. Requires a confirmation. 'confirmation' must read literally 'CANCEL <job_id>'.
check_isolation
Cross-check of the isolation: does the tunnel work, and is the LAN really closed. Checks by the real behavior, not the configuration: once through the tunnel, once directly against the LAN address. After isolation the first path must work and the second must fail.
check_training_env
Whether the LoRA environment on the VM is present and actually importable.
create_variant
Creates a custom variant from an existing Ollama model. The variant gets a fixed system prompt and its own parameters and is afterwards addressable under 'name' like any other model, also through a separate query bridge. name Lowercase letters, digits, dot, hyphen, underscore, optionally a tag after a colon, e.g. 'handyman:v1'. An already-taken name is rejected, unless it is a variant created here; that one is then replaced. base_model must be in the Ollama inventory on the VM. If it is missing, this tool says so…
fetch_model
Fetches a model onto the VM. Returns immediately, with an id. The fetch takes minutes to hours depending on size and therefore does not block this call. Progress comes from 'job_status' for the returned id. source 'ollama' fetches into the Ollama inventory, source 'huggingface' into the cache that vLLM loads from. Free disk space is checked beforehand; if it does not fit, the fetch is rejected rather than filling the disk.
job_status
Progress of a background job, e.g. a model fetch.
list_adapters
Which finished LoRA adapters are on the VM, for which base model.
list_jobs
All jobs known in this session plus the folders on the VM.
list_variants
Which variants exist, where they come from and what is set in them. Additionally names the space usage of the Ollama store and the list of tunable parameters with their ranges.
models_present
Which models are on the VM, with size and storage location. Split by the two inventories: the models managed by Ollama and the ones in the Hugging Face cache that vLLM loads from.
remove_model
Removes a model from the VM. Requires an explicit confirmation. 'confirmation' must read literally 'REMOVE <model name>', with exactly the name that is also in 'model'. This keeps a model from disappearing by accident.
remove_variant
Removes a variant. Requires an explicit confirmation. 'confirmation' must read literally 'REMOVE <name>', with exactly the name that is also in 'name'. Removed are the variant in Ollama and the associated Modelfile on the VM. The base model stays untouched.
show_variant
What a variant really contains. Shows both side by side: what Ollama has stored for the name, and the Modelfile this tool wrote for it. If the two diverge, the variant was changed by hand.
start_service
Starts a previously halted service again and checks whether it answers.
start_training
Starts a LoRA training run in the background and returns an id. Important: inference and training cannot share the card. If the inference service is running, it is stopped first. This does not happen silently: as long as 'stop_inference_confirmed' is not True, the start is rejected and it is instead reported what would be stopped. 'data_path' must point to a JSONL file beneath the configured data root (MM_VM_DATA_ROOT) on the VM. The run outlives the end of this session; progress with 'training_status'.

Show 4 more tools

status
What is currently running on the model machine. Reports reachability, the inference service and served model, loaded Ollama models, GPU memory usage and free disk space. If the VM is unreachable, that is reported as a state of its own: the card is then most likely assigned to the other VM, and that is not an error.
stop_service
Halts one of the two services. 'vllm' or 'ollama'.
switch_service
Switches the inference service to a different model. Before the switch it is checked whether the model is on the VM and whether it fits into GPU memory with the desired context; the estimate is reported. If it does not fit by calculation, it is rejected rather than tried blindly. Image, mounts and ports are read from the running container, not assumed. After the restart it is checked whether the service really answers and serves the desired model. A switch that fails silently is the most dangerous failure mode; that is why the result of this cross-check is put in the report for the normal ou…
training_status
Progress, loss and estimated time remaining of a training run.

Signals

Stars: 0
Forks: 0
Open issues: 0
Weekly downloads: 249
Last commit (this folder): Oct 22, 2026 (10 days ago)
Last release: Oct 22, 2026 (10 days ago)
Last package publish: Oct 22, 2026 (10 days ago)

From GitHub (fetched Oct 24, 2026) and the package registry.

Details

Registry name: io.github.setheerwagen/mcp-modelmanager
Version: 1.0.6
Auth: No auth
Packages: pypi mcp-modelmanager
License: MIT
Repository: github.com/setheerwagen/local-ai-mcp-servers
First seen: Oct 24, 2026
Last checked: Oct 24, 2026
Source: Official MCP Registry github-graphql

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped: 50+; launches shipped
media placements: 500+; media placements
founders on every account: 2; founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get Model Manager running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Need help deploying Model Manager?

Call Get help

Cite this page

Model Manager MCP server. Agent Mart. https://agentmart.dev/connectors/setheerwagen-mcp-modelmanager (last updated 2026-10-24)

Need help deploying Model Manager?

Install

About

Reviews

Permissions

Tools by what they do 19

Read 10 · Look things up without changing anything.

Write 5 · Create, change, send or run things.

Delete 2 · Delete, drop, revoke or wipe things.

Not yet classified 2 · We could not tell from the name and description.

Secrets and settings it asks for

Network hosts it talks to 1

Files it touches

Tools (19)

Signals

Details

More from setheerwagen

Rather not set it up yourself?

Get Model Manager running in your business