Skip to main content
Glama
Sign In
llm-wiki-mcp

llm-wiki-mcp

by quick-sort
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

llm-wiki-mcp

HTTP MCP server providing wiki knowledge base tools (query and ingest), powered by llm-wiki-agent skills via the Claude Agent SDK.

Architecture

MCP Client (Claude Code, etc.)
    ↓ HTTP / stdio
llm-wiki-mcp (FastMCP server)
    ↓ Claude Agent SDK (query())
Claude Code agent session
    ↓ auto-loaded skills
llm-wiki-agent (.claude/commands/)
    ↓ /wiki-ingest, /wiki-query
Wiki markdown files (entities, concepts, sources, syntheses)

The MCP server is a thin layer. The actual knowledge extraction, entity/concept page creation, cross-referencing, and contradiction detection are all handled by the llm-wiki-agent skill running inside a Claude Agent SDK session.

Related MCP server: mcp-wikijs

Quick Start

# Install dependencies
uv sync

# Clone llm-wiki-agent skills
git clone --depth 1 https://github.com/SamurAIGPT/llm-wiki-agent.git skills/llm-wiki-agent

# Set your Anthropic API key
export ANTHROPIC_API_KEY=sk-ant-...

# Start the server (HTTP mode, default port 8080)
uv run llm-wiki-mcp

# Or use stdio mode
uv run llm-wiki-mcp --transport stdio

Tools

ingest

Ingest a document (file path or URL) into a named wiki. The llm-wiki-agent skill handles reading, converting, extracting knowledge, and building cross-referenced pages.

ingest(wiki_name="my-research", source="raw/papers/attention-is-all-you-need.md")
ingest(wiki_name="my-research", source="report.pdf")

query

Query a wiki with a natural language question. The skill searches all wiki pages and synthesizes an answer with [[wikilinks]].

query(wiki_name="my-research", question="What are the main approaches to attention?")

Docker

docker build -t llm-wiki-mcp .
docker run -e ANTHROPIC_API_KEY=sk-ant-... -p 8080:8080 -v ./wikis:/home/agent/wikis llm-wiki-mcp

Optional: configure custom base URL or models:

docker run \
  -e ANTHROPIC_API_KEY=sk-ant-... \
  -e ANTHROPIC_BASE_URL=https://custom.anthropic.com \
  -e ANTHROPIC_SMALL_FAST_MODEL=claude-haiku-4-20250514 \
  -e ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-4-20250514 \
  -e ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-4-20250514 \
  -e ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-haiku-4-20250514 \
  -p 8080:8080 \
  -v ./wikis:/home/agent/wikis \
  llm-wiki-mcp

Available Tools

2 tools
ingestA

Ingest a document into a wiki knowledge base.

The source is fetched (URL or local path) and saved to the wiki's raw/ directory, then the llm-wiki-agent /wiki-ingest skill processes it — extracting entities, concepts, and cross-references into structured wiki pages.

Args: wiki_name: Name of the wiki to ingest into (created if it doesn't exist). source: A local file path or URL to ingest.

Inputs: wiki_name (required), source (required). Returns: result

queryA

Query a wiki knowledge base and get an AI-synthesized answer.

Uses the llm-wiki-agent /wiki-query skill to search wiki pages and synthesize a comprehensive answer with [[wikilinks]] citations.

Args: wiki_name: Name of the wiki to query. question: The question to ask about the wiki's knowledge.

Inputs: wiki_name (required), question (required). Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A3.9/5.0

    Scored across 2 tools

    Disambiguation5/5

    The two tools, 'ingest' and 'query', have clearly distinct purposes: one adds documents to the wiki, the other retrieves information. There is no ambiguity or overlap in their functionality.

    Naming Consistency5/5

    Both tool names follow a single-verb pattern that directly describes their action ('ingest' and 'query'), making them consistent and predictable.

    Tool Count3/5

    With only two tools, the server feels under-scoped for a wiki knowledge base. While the tools cover core ingest and query actions, a more complete server would include additional management tools.

    Completeness2/5

    The server lacks essential operations such as listing wikis, deleting wikis, or updating ingested content. Users have no way to manage the wiki beyond ingesting and querying, which are significant gaps.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Resources

    Related MCP Connectors

    Related MCP Servers