MarkNest MCP Server

Stdio MCP server for connecting AI assistants to the MarkNest public V1 API.

Current coordinated release: 0.2.0.

This package intentionally uses only documented public endpoints under /api/v1. It does not call internal app endpoints for writes.

Tools

• marknest_list_documents
• marknest_get_document
• marknest_create_document
• marknest_save_summary
• marknest_update_document
• marknest_delete_document
• marknest_list_document_shares
• marknest_get_document_share
• marknest_get_document_share_for_document
• marknest_create_document_share
• marknest_update_document_share
• marknest_delete_document_share
• marknest_list_document_versions
• marknest_get_document_version
• marknest_list_folders
• marknest_get_folder
• marknest_create_folder
• marknest_update_folder
• marknest_delete_folder
• marknest_list_folder_documents
• marknest_list_tags
• marknest_create_tag
• marknest_update_tag
• marknest_merge_tag
• marknest_delete_tag
• marknest_list_tag_documents
• marknest_list_files
• marknest_get_file
• marknest_download_file
• marknest_upload_file
• marknest_update_file
• marknest_delete_file
• marknest_search_chunks — semantic search across document chunks (RAG retrieval; requires rag:read)
• marknest_get_chunk — fetch one full chunk by ID (requires rag:read)

Document Listing

marknest_list_documents returns document summaries by default. Summary rows omit content and rendered_html to match the public V1 API list behavior and keep MCP responses small.

Use these options when needed:

• include_content: true appends include=content and returns content plus rendered_html in list rows.
• full_text_search: true appends search_scope=full_text; without it, search matches titles only.
• page and per_page control pagination.

Prefer marknest_get_document when the client needs one full document body.

Configuration

Set these environment variables in your MCP client config:

MARKNEST_BASE_URL=https://api.marknest.net
MARKNEST_API_KEY=mk_your_public_api_key

Create the API key from MarkNest dashboard Settings -> API Keys. Grant the abilities needed by the tools you want the client to use:

• documents:read
• documents:write
• folders:read
• folders:write
• tags:read
• tags:write
• files:read
• files:write
• rag:read — required by marknest_search_chunks and marknest_get_chunk
• shares:read, shares:write — required by document-share tools

Safety-sensitive tools

• marknest_merge_tag is irreversible and requires explicit source and target tag IDs.
• Share creation or access-policy changes may expose content. The caller must explicitly identify the document and access policy.
• marknest_download_file requires an explicit output_path; it refuses to overwrite an existing file unless overwrite: true is supplied. Use disposition: inline to call the inline endpoint.

Semantic Search (RAG)

marknest_search_chunks runs hybrid vector + keyword retrieval over the user's knowledge base and returns chunk excerpts with heading_path breadcrumbs and document slugs for citation. marknest_get_chunk fetches the full chunk text when the excerpt is not enough. The API key must have the rag:read ability; both tools call POST /api/v1/search/semantic and GET /api/v1/documents/{document_id}/chunks/{chunk_id}.

These tools let an AI assistant (Claude Desktop, ChatGPT, etc.) answer questions from a user's MarkNest knowledge base without MarkNest itself hosting an LLM — the assistant receives the chunks and constructs the answer.

Run

Via npx (no local checkout needed):

npx @marknest/[email protected]

Or from this package:

pnpm --filter @marknest/mcp-server start

Example MCP client entry:

{
  "mcpServers": {
    "marknest": {
      "command": "npx",
      "args": ["-y", "@marknest/[email protected]"],
      "env": {
        "MARKNEST_BASE_URL": "https://api.marknest.net",
        "MARKNEST_API_KEY": "mk_your_public_api_key"
      }
    }
  }
}

MARKNEST_BASE_URL defaults to https://api.marknest.net and can be omitted unless you target a different deployment.

Smoke Test

pnpm --filter @marknest/mcp-server smoke