MCP Wiki Server
Allows the server to source a wiki knowledge base from a remote git repository: it shallow-clones the repo on first run and refreshes it with git pull --ff-only on subsequent starts, serving the cached copy if a refresh fails.
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., "@MCP Wiki Server show me the table of contents for the sql wiki"
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.
MCP Wiki Server
An MCP server that exposes a wiki / knowledge base to AI agents. Each top-level folder under the wiki root becomes its own tool, so the agent can see the available domains directly from the tool list and pull only what it needs into context.
wiki/sql/ -> tool wiki_sql
wiki/python/ -> tool wiki_python
wiki/git/ -> tool wiki_gitTool semantics
Every topic tool accepts the same two optional arguments:
call | returns |
| Table of contents for the topic (filename + summary) |
| Matching lines across all pages with |
| Full page contents (truncated at 8 KB) |
Related MCP server: wikikit
Configuration
The server reads its wiki source from environment variables.
Local folder (default)
WIKI_PATH=/absolute/path/to/wikiIf unset, defaults to ./wiki next to server.py (the bundled demo).
Remote git repo
WIKI_GIT_URL=https://github.com/<org>/<wiki-repo>
WIKI_GIT_BRANCH=main # optional, defaults to "main"
WIKI_CACHE_DIR=/tmp/my-wiki # optional, defaults to system tempdirWIKI_GIT_URL takes precedence over WIKI_PATH. The repo is cloned shallowly
on first run and refreshed with git pull --ff-only on every start after that,
so a long-lived cache does not serve the wiki frozen at whenever it was first
cloned. A failed refresh (offline, revoked credentials) is reported on stderr
and the cached copy is served as-is; delete the cache dir to force a fresh
clone.
Running
MCP Inspector (for testing)
# The <2 bound matches pyproject.toml — server.py uses the v1 FastMCP API,
# which mcp 2.x renamed. Without it, --with resolves to 2.x and the import fails.
uv run --with "mcp[cli]<2" mcp dev server.pyOpen the printed URL → confirm wiki_sql, wiki_python, wiki_git appear and
call them with the argument shapes above.
From Claude Code
Copy .mcp.json.example into your project as .mcp.json (or merge it into
~/.claude.json for global use), restart Claude Code, then run /mcp to
verify the connection.
Relationship to the skills in this repo
None. The architecture and refactoring plugins used to embed this server
over their skills' references/ folders; that wiring was removed because it
only duplicated what an agent gets by reading those pages directly. The
skills now rely on plain progressive disclosure, and this server stands on
its own for wikis that live outside a skill.
The server follows directory symlinks, so topic folders under WIKI_PATH
may themselves be links.
Adding your own content
Create a folder under your WIKI_PATH and drop Markdown files in it:
my-wiki/
kubernetes/
deployments.md
services.md
terraform/
modules.mdRestart the server. New tools wiki_kubernetes and wiki_terraform will
appear automatically. Folder names with non-identifier characters are
sanitized: data-science/ becomes wiki_data_science.
Sanitizing maps every non-alphanumeric character to _, so sibling folders can
collapse onto one tool name — sql-1/ and sql_1/ both become wiki_sql_1.
The server registers the first and skips the rest, naming the collision on
stderr; rename one of them.
Limits
Pages are truncated at 8000 characters per fetch.
Search returns at most 20 hits per call.
Path traversal via
page=../...is blocked.Only
*.mdfiles are indexed.
Available Tools
3 toolswiki_gitA
Search or read the 'git' wiki topic. No args: list pages. query=: find matches. page=<filename.md>: full page.
Inputs: page, query. Returns: result
wiki_pythonA
Search or read the 'python' wiki topic. No args: list pages. query=: find matches. page=<filename.md>: full page.
Inputs: page, query. Returns: result
wiki_sqlA
Search or read the 'sql' wiki topic. No args: list pages. query=: find matches. page=<filename.md>: full page.
Inputs: page, query. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Tools are clearly separated by wiki topic (git, python, sql), so an agent can pick the right one when the topic is known. However, the identical interface and absence of a generic topic parameter or cross-topic search mean ambiguity arises when the topic is unspecified or spans multiple wikis.
All tool names follow a consistent wiki_<topic> pattern (wiki_git, wiki_python, wiki_sql), making them predictable and easy to parse.
Three tools map to three topics, which is a reasonable number for the stated scope, though a single parameterized tool would reduce duplication and scale better.
Read/list/search operations are covered per topic, but there is no write capability (create/edit/delete) and no way to discover or query other topics, leaving notable gaps for a wiki server.
Maintenance
Related MCP Connectors
Markdown workspace for AI agents: read, write, organize, and share markdown documents.
Shared project wiki for AI agents: read and write pages, next actions, and activity logs over MCP.
Markdown notes and whiteboards your AI agent can read and write.
Shared knowledge base for AI agents. Semantic search across agents, no setup required — just a URL.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to search, read, and traverse a local knowledge base of Markdown files using full-text search and relationship graph, reducing token usage.MIT
- AlicenseAqualityDmaintenanceEnables agents to search, list, and read pages from a curated Markdown wiki via MCP, with live updates and tools for targeted section access.5MIT
- AlicenseNot gradedqualityBmaintenanceEnables agents to semantically search and read curated local markdown knowledge bases through MCP, with optional CLI and dashboard tooling.MIT
- AlicenseNot gradedqualityNot gradedmaintenanceEnables any MCP-capable agent to write and maintain its own wiki of small linked markdown pages with a typed ontology, storing, indexing and retrieving them through keyword (BM25) search plus ontology-graph traversal, typed-relation management, page reads and writes, linting and listing. It also offers an optional vision tool that extracts objects and text from images so text-only models can learn from screenshots, charts and scans.MIT