Offline Mcp

io.github.gabrielmahiaPublished by GitHub account gabrielmahiav0.1.13Updated Nov 11, 2026

Local AI over Ollama for low connectivity: status, local inference, model and deployment guides.

VerifiedSTDIODesktop onlyCloud & InfrastructureAI & ML

Overview

AI-generated overview

Wraps a local Ollama install so an assistant can check status, run prompts on local models, and get model and deployment guidance.

What you can do
  • Check whether Ollama is running locally and list the models it has
  • Run a prompt through a local Ollama model
  • List recommended open-weight models for low-connectivity use cases
  • Read guides on degraded-mode operation, open-weight model options, and deployment on laptop, server, Raspberry Pi or Android
When to use it
Useful when you want AI assistance to run entirely on the local machine, without cloud APIs or API keys, especially on constrained hardware or intermittent connectivity.
Requirements
Python package installed with pip; a local Ollama installation with models pulled for inference to work. No API keys, environment variables or headers are declared.
Before you install
The README states demo data is labeled DEMO and is not suitable for operational decisions. Inference quality and availability depend on the local Ollama setup and hardware; the README makes broad claims about data sovereignty that you should verify yourself.

Installation

In SourceWeft

  1. Open Offline Mcp in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

offline-mcp

[offline-mcp Glama score]

Install: pip install offline-mcp · Use with any MCP client.

Local AI inference infrastructure — Ollama wrapper, open weights directory, degraded-mode guide for East Africa.

[PyPI] [Thesis Layer]

Why: Never assume OpenAI survives, Anthropic stays accessible, or export controls disappear. This matters more in Africa than anywhere else.

1st world equivalent: Ollama, LLaMA, Mistral local deployment

Why This Exists: Data Sovereignty

"If you take the deal, you're going to be exploited. If you don't take it, you're going to die."
— Frank Ssekamwa, Ugandan digital rights expert

Across the Global South, AI and health data from communities is being extracted, processed abroad, and used to build models whose value flows away from the communities that generated it.

offline-mcp is the sovereignty floor of the East Africa coordination stack.

When this runs on a Raspberry Pi 4 with a 50W solar panel and a 256GB SD card:

  • Health data stays in the clinic. Health guidance comes from local models.
  • Land records stay in the land office. Queries don't touch foreign servers.
  • Civic data stays in the county. AI assistance runs without internet.

No API key. No cloud dependency. No data leaving the community.

The models available via offline-mcp (Llama 3.2, Qwen 2.5) run entirely on device. Community data used to generate AI outputs creates no dataset sent back to model providers.

This is not a privacy feature. It is the architectural foundation of digital independence.


Research Foundation

This server implements patterns validated by peer-reviewed research on offline-first AI for bandwidth-constrained environments:

arXiv:2603.03339 (2026) — Offline-First LLM Architecture for Adaptive Learning in Low-Connectivity Environments — confirms that meaningful AI support is achievable with hardware-aware model selection when designed for infrastructure-limited deployment. Key finding: offline-first is a complementary paradigm, not a compromise.

Design principles applied:

  • Local-first: all core operations execute without internet connectivity
  • Graceful degradation: reduced functionality beats no functionality
  • JSONL queue: events accumulated offline sync when connectivity returns
  • Hardware-aware: tool selection adapts to available compute

East Africa context: Kenya, Tanzania, Uganda — rural areas with intermittent connectivity represent the primary deployment target. This server is not designed for ideal conditions. It is designed for real ones.

Install

bash
pip install offline-mcp

Tools (6)

ToolDescription
check_ollama_statusCheck if Ollama is running locally and list available models
run_local_inferenceRun a prompt through a local Ollama model
list_recommended_modelsBest open-weight models for East Africa use cases
degraded_mode_guide4-level degraded mode architecture for offline operation
open_weights_directoryDirectory of open-weight models with Africa language support
local_deployment_guideDeployment guide for laptop, server, Raspberry Pi, Android

Context

Runs on a 50W solar panel + Raspberry Pi 4. Viable for rural Kenya clinics, schools, and community offices.

→ The Nairobi Stack

License

MIT © Gabriel Mahia | [email protected]

Part of the East Africa Coordination Stack

This MCP server is part of the Kenya coordination infrastructure. Connect it to africa-coord-bus — the coordination event bus that routes signals between domains automatically.

bash
pip install africa-coord-bus

All servers: pypi.org/user/gmahia Live demo: coord-cascade-demo

IP & Collaboration

MIT licensed. Feedback via GitHub Issues only — pull requests are not accepted. Demo data is labeled DEMO and is not suitable for operational decisions. Full policy: docs/architecture/IP_POLICY.md. Security reports: see SECURITY.md.

Part of the East Africa coordination stack

Model-agnostic by design: closed APIs, open-weight models, and small distilled models are all first-class citizens.

Source: README.md at commit 124aa0e

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

9
  1. v0.1.13LatestNov 11, 2026
  2. v0.1.12Nov 7, 2026
  3. v0.1.11Nov 3, 2026
  4. v0.1.10Oct 23, 2026
  5. v0.1.9Oct 21, 2026
  6. v0.1.8Oct 17, 2026
  7. v0.1.7Oct 13, 2026
  8. v0.1.5Oct 12, 2026
  9. v0.1.4Sep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts