ollama-mcp-server
Provides a tool to delegate prompts to Ollama models, sending text generation requests to the Ollama API with an optional system prompt and configurable model.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ollama-mcp-serveruse qwen3.5:4b to summarize this 2000-line log file"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ollama-mcp-server
A thin TypeScript MCP server for delegating bounded work to Ollama and for
inspecting repositories without spending the host agent's context on bulk work.
It runs over stdio with npm start; there is no build step.
For a concise orientation to the repository layout and the purpose of each tracked file, start with the project map.
Supported now
The default MCP surface is intentionally small.
Basic Ollama tools
run_ollama_tasksends one stateless task to an Ollama model.summarize_outputcondenses large text through Ollama.list_ollama_modelslists available local or signed-in models.
Quality Review
The separate read-only CLI scans JavaScript/TypeScript functions into SQLite,
reviews a bounded queue, and saves reports under the target repository's
.quality-review/ directory. It never changes target source files.
npm run quality -- scan --cwd /path/to/repo
npm run quality -- review --count 10 --cwd /path/to/repo
npm run quality -- status --cwd /path/to/repoFindings can later be inspected and marked accepted or rejected; those decisions are metadata only. See the Quality Review guide.
Lean Explorer
The default Explorer uses deterministic repository intelligence:
outline_fileandread_symbolfind_symbol,find_references,find_callers, andfind_calleeshybrid_retrieveinbasicmode, combining lexical and structural ranking with relevant documentation and package scripts
basic mode does not use embeddings or autonomous model planning. A client can
retrieve likely symbols, read focused source, and summarize the evidence. Use
mode: "lexical" for the narrow symbol baseline. mode: "hybrid" is available
only when semantic search and Git-history intelligence are enabled.
The supported implementation is isolated in src/explorer/. Advanced and
autonomous code is parked under src/experimental/ and is not imported during
default startup.
Deterministic CLI entry points remain available, for example:
npm run index:super-explorer -- /path/to/repo
npm run hybrid:super-explorer -- basic /path/to/repo "where is config loaded"Related MCP server: Ollama MCP Server
Experimental and disabled by default
Experimental code remains in the repository, but its MCP tools are not
registered unless enabled. ENABLE_EXPERIMENTAL=1 enables all experimental
groups; a group-specific 0 can override the master flag.
Flag | Enables |
|
|
| Git-history CLI/ranker; the other prerequisite for full |
|
|
|
|
|
|
|
|
| framework-adapter CLI entry points |
local_explore_repo is the preferred model-backed scout. It retrieves up to
8–12 basic candidates, adds bounded caller and source-text context, and sends
excerpts from at most six files to qwen3.5:4b by default. The model has no
tools or shell access in this route. It selects candidate IDs and exact source
quotes; the server checks those quotes and retries once if they fail. The parent
agent interprets the evidence—quote checking cannot prove a behavioral claim.
The older local_explorer_task loop remains for historical comparisons.
The corresponding semantic, knowledge, verification, full-Explorer, and
framework CLI commands enforce the same gates. The full pipeline defaults to
deterministic basic retrieval; full hybrid mode additionally requires both
ENABLE_SEMANTIC_SEARCH=1 and ENABLE_GIT_HISTORY=1.
Autonomous workers
Git-writing workers are a separate safety category and are never enabled by
ENABLE_EXPERIMENTAL:
LOCAL_WORKER_ENABLED=1registersrun_local_worker_task.CLOUD_CLAUDE_ENABLED=1registersrun_cloud_claude_task.
Both retain the shared Git-command allowlist and cloud validation hook. Verify
their work with git status and git log; do not trust a worker's own report.
Frozen research
Historical benchmark results, model-routing experiments, fine-tuning plans,
framework research, and autonomous-agent experiments are preserved under
docs/ and scripts/. They are not being expanded and are not part of the
normal runtime path. Start with the documentation index.
Setup
Requires Node 22.13+ and an Ollama server for model-backed tools.
npm install
ollama serve
npm startCopy .env.example to .env only when configuration overrides are needed.
Register npm start as a local stdio MCP server in your client.
Verification
npm run test:features
npm run test:quality
npm run test:local-explore
npx tsc --noEmitThe Quality Review integration test binds a temporary localhost port for a mocked Ollama endpoint.
This server cannot be deployed
Maintenance
Related MCP Connectors
Agent personas for Claude. 16 tools, 13 personas, 3 workflows. Zero extra API cost. Free.
Use your own Mac from ChatGPT, Claude or Codex: files, commands, documents, and a browser.
Live SEO workflow tools for Claude Code, Codex, and AI agents.
Source-checked CLI guides and model-aware planning for Claude Code, Codex, and Grok Build.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables Claude to delegate coding tasks to local Ollama models, reducing API token usage by up to 98.75% while leveraging local compute resources. Supports code generation, review, refactoring, and file analysis with Claude providing oversight and quality assurance.479 npm25AGPL 3.0
- AlicenseNot gradedqualityDmaintenanceExposes local Ollama instances as tools for Claude Code, allowing users to offload code generation, text drafting, and embedding tasks to local GPUs. It supports multi-turn conversations and model management through the Model Context Protocol.MIT
- AlicenseNot gradedqualityCmaintenanceEnables Claude Code to delegate mechanical tasks (summaries, boilerplate, reformatting) to local models running in LM Studio.1MIT
- AlicenseAqualityCmaintenanceEnables Claude Code to offload routine code generation and text processing tasks to a local Ollama LLM, saving Cloud API tokens and costs with automatic model selection and security features.1163 npm4Apache 2.0