M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
L

llmwiki MCP Server

Community
Compile raw sources into an auditable, interlinked knowledge base.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #158 of 438
Stars
★ 2.2k Very popular
Runtime
Node.js 24+
Credentials
Optional API key
License
MIT
Last commit
Tools
2
57FMRS · C

The source identifies an MCP server and two explicitly named MCP tools, get_context_pack and run_eval, but does not specify transport details, client compatibility, or client installation configuration. The project is MIT-licensed and uses the repository root as its server path. Its primary purpose is compiling source material into reviewable, citation-aware, searchable knowledge.

Strongest · Documentation 15/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
12/20
Maintenance
12/20
Documentation
15/20
Setup experience
10/20
Why each score
Reliability 8/20
The README describes the MCP server, tool scope, providers, and fail-closed behavior, but the server manifest is unavailable and there is no actual MCP initialization/tool-list response, source code, or committed server-test evidence. CI and test commands are only README claims; static review cannot verify key paths, error handling, or dependency control, so the score reflects plausible but unverified operation.
Security and permissions 12/20
The README claims project-path confinement, call-time credential validation, review staging for external imports, artifact hash checks, size limits, and a read-only viewer, indicating a reasonably explicit security model. Points are deducted because there is no auditable implementation, tool-level permission matrix, MCP client authorization configuration, network-flow detail, or evidence of per-operation confirmation for risky writes. No explicit red-line risk is shown in the supplied material.
Maintenance 12/20
The repository is MIT-licensed and not archived; the README claims releases 1.1.0 and 1.0.0 and references CI, contribution guidance, and 15 open issues, indicating ongoing project activity. Points are deducted because the evidence is README-only: commit history, release cadence, dependency updates, security-response channel, and issue-response quality are not provided. Stars do not add points.
Documentation 15/20
The README covers scope, commands, MCP startup, SDK use, Node.js requirements, several providers, environment variables, import/export, quality controls, limitations, and documentation entry points, with concrete examples. Points are deducted for missing MCP tool parameter/return documentation, mainstream-client connection examples, cost and rate-limit disclosure, troubleshooting evidence, and cross-validation against a manifest or source files.
Setup experience 10/20
The README provides a basic path using Node.js 24, global npm installation, environment variables, project initialization, and `llmwiki serve --root`; it also identifies some credential-free capabilities. Points are deducted because no mainstream MCP client configuration is shown, the Node.js 24 requirement is relatively demanding, provider and credential choices add configuration complexity, and there is no static evidence of a working handshake or cross-platform validation.

Static review · not runListed 2026-08-14

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesUses the network

Best for

  • Research material that benefits from review, provenance, and reuse
  • Codebase documentation, team handbooks, standards, design notes, and decision logs
  • Agent workflows that precompile knowledge before querying it

Not for

  • High-churn log streams where raw search is sufficient
  • General-purpose static-site generation
  • Heavy ontology databases

Required permissions

  • Access to the selected knowledge-base project and its sources, wiki, and .llmwiki data
  • Appropriate network access when ingesting web pages or using external connectors; the README does not specify connector permissions
  • Provider credentials at call time for LLM-backed operations

Risks and side effects

  • LLM-generated content may contain confidence, contradiction, provenance, or freshness issues
  • External OKF imports enter the review queue by default, but explicitly trusted imports can write to the active knowledge base
  • Ingest, connector, import, and other write operations can change project data; review, lint, and eval should be used
  • The README does not specify an MCP transport or client configuration, so client compatibility cannot be assumed

Setup

Before you start

Runtime：Node.js 24+

ANTHROPIC_API_KEY optionalsecret Anthropic API key, the default provider credential used by LLM-backed tools; get it from the Anthropic console.
ANTHROPIC_AUTH_TOKEN optionalsecret Alternative Anthropic credential for ANTHROPIC_API_KEY.
OPENAI_API_KEY optionalsecret API key for OpenAI-compatible providers, needed when LLMWIKI_PROVIDER=openai.
GITHUB_TOKEN optionalsecret Token for the GitHub Copilot provider, typically set to $(gh auth token).
Other optional settings (4)
LLMWIKI_PROVIDER optional Selects the LLM provider (e.g. openai, ollama, copilot, claude-agent); defaults to anthropic if unset.
OPENAI_BASE_URL optional Optional base URL for an OpenAI-compatible endpoint.
OLLAMA_HOST optional Host address of a local Ollama instance, used when LLMWIKI_PROVIDER=ollama.
LLMWIKI_COMPILE_CONCURRENCY optional Optional cap on parallel compile concurrency.

Node.js 24 or newer is required. Install with npm install -g llm-wiki-compiler, then run llmwiki serve --root <project-directory> from the knowledge-base project.

Check that it works

Start the server with `llmwiki serve --root /path/to/wiki-project`, then confirm get_context_pack and run_eval appear in the client's tool list; calling run_eval (fast suite, no API key needed) and getting a health score back proves the connection works.

Troubleshooting

  1. Verify that Node.js is version 24 or newer
  2. Verify that --root points to the knowledge-base project directory
  3. Run llmwiki status, llmwiki lint, and llmwiki eval to inspect state and quality
  4. If an LLM-backed operation fails, check the selected provider and its credentials
  5. For stale pages, run llmwiki refresh --stale --dry-run before applying repairs

Things to try

Once connected, you can ask your AI assistant things like:

  • Build a context pack for the task of summarizing this project's core concepts
  • Run the fast eval suite and report the wiki's health score
  • Run a full eval on my research wiki to check citation coverage
  • Get a citation-aware context pack I can hand to another agent

Tools 2

get_context_pack read-only
Builds a citation-aware evidence context pack for an agent.
run_eval writes
Runs knowledge-base quality evaluation; the fast suite needs no provider, while the full suite's citation-support judging requires one.

Use cases

Compile raw material into a persistent interlinked Markdown wiki
Build citation-aware evidence packs for agents
Check citations, links, stale pages, and knowledge quality
Query, read, and exchange compiled knowledge through MCP

Overview

llmwiki is a knowledge compiler that turns papers, notes, READMEs, transcripts, PDFs, images, or web pages into an interlinked Markdown wiki with citations, metadata, review state, and provenance. Its MCP server lets compatible agents ingest, compile, query, read, lint, evaluate, inspect status, build context packs, and exchange Open Knowledge Format bundles. The server path is the repository root.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 50380af5558a Data synced 2026-10-19 Read the FMRS scoring method