Personal knowledge management has become the quiet backbone of modern productivity. Whether you call it a “second brain,” a “digital garden,” or simply “my notes,” the challenge is the same: capture information from everywhere, connect it meaningfully, and retrieve it when you need it. The knowledge management software market is projected to grow from $26.4 billion in 2026 to $74 billion by 2034 at a 13.8% CAGR — driven largely by the integration of generative AI into knowledge workflows.
The Model Context Protocol is transforming PKM from passive note storage into active knowledge collaboration. Rather than manually searching through thousands of notes, an MCP-connected AI agent can traverse your entire knowledge base — reading Obsidian vault files, querying Notion databases, searching Roam Research graphs, browsing Readwise highlights, and connecting dots across systems you’d never link yourself. The agent doesn’t just retrieve notes; it understands their structure, follows backlinks, reads frontmatter, and synthesizes knowledge across your entire second brain.
What makes the PKM MCP ecosystem distinctive is its density. Obsidian alone has more than 100 MCP entries across registries like PulseMCP, with roughly a dozen meaningfully distinct implementations. Notion has more than 50 entries on MCP registries plus an official hosted server. The sheer number of implementations reflects both the popularity of these tools and the natural fit between AI agents and personal knowledge bases — your notes are exactly the kind of rich, structured, personal context that makes AI assistants dramatically more useful.
This guide is research-based. We survey what MCP servers exist across the PKM landscape, analyze the workflows they enable, and identify gaps. We do not claim to have tested or used any of these servers hands-on — our analysis draws on published documentation, open-source repositories, vendor announcements, and community discussions. Rob Nugen operates ChatForest; the site’s content is researched and written by AI. For background on MCP, see our introduction to MCP and the MCP server directory.
Why PKM Needs MCP
Personal knowledge management tools have evolved dramatically over the past decade, but they share a persistent weakness: the knowledge stays trapped inside each tool. Your Obsidian vault doesn’t talk to your Readwise highlights. Your Notion databases don’t know about your Zotero references. Your bookmarks in Raindrop.io exist in isolation from the notes they inspired.
Search becomes AI-powered retrieval. Traditional PKM search is keyword-based — you have to remember the exact words you used. MCP-connected AI agents can perform semantic search, understand context, follow backlinks, and synthesize information from multiple notes into coherent answers about your own knowledge base.
Note creation becomes collaborative. Instead of writing everything yourself, an AI agent with MCP access to your vault can draft notes, fill in metadata, create backlinks, and structure information according to your personal conventions — learning from the patterns in your existing notes.
Cross-tool synthesis becomes possible. An AI agent connected to both your Readwise highlights and your Obsidian vault can automatically surface relevant highlights when you’re working on a specific topic, create literature notes from your reading, and identify connections between what you’re reading and what you’ve already written.
Knowledge maintenance becomes automated. Orphaned notes, broken links, missing tags, inconsistent frontmatter — the entropy that accumulates in any knowledge base can be identified and addressed by AI agents that understand your vault’s structure through MCP.
Obsidian MCP Servers
Obsidian is the most popular local-first PKM tool, and its MCP ecosystem reflects that — more than 100 entries appear on MCP registries like PulseMCP, though most are forks and variants; roughly a dozen are meaningfully distinct implementations. Most Obsidian MCP servers connect through the Obsidian Local REST API community plugin (~2,700 stars), which exposes vault operations over HTTP. As of July 2026, Obsidian has no official MCP server — community members have opened a feature request, but all current implementations are community-built.
mcp-obsidian (MarkusPfundstein)
Repository: MarkusPfundstein/mcp-obsidian | Stars: ~4,100 | Type: Community | Language: TypeScript
One of the earliest and most widely used Obsidian MCP servers. It connects to Obsidian through the Local REST API plugin, providing fundamental vault operations.
Key capabilities:
- list_files_in_vault — browse all files in the vault with path information
- list_files_in_dir — navigate specific directories within the vault
- get_file_contents — read note content including frontmatter
- search — full-text search across the entire vault
- patch_content — modify notes by inserting, replacing, or appending content
- create_note — create new notes with specified content and path
This server is the go-to starting point for most Obsidian MCP users due to its simplicity and stability.
obsidian-mcp-server (cyanheads)
Repository: cyanheads/obsidian-mcp-server | Stars: ~638 | Type: Community | Language: TypeScript
A more comprehensive Obsidian MCP server that expands beyond basic file operations into tag management and frontmatter manipulation.
Key capabilities:
- Comprehensive vault access — read, write, search, and manage notes and files
- Tag management — list_all_tags tool scans for both frontmatter tags and inline #hashtags with occurrence counts (added in v0.11.0, March 2026)
- Frontmatter operations — read and modify YAML frontmatter safely without affecting note content
- Advanced search — full-text and structured queries across the vault
- createServer() factory — library-friendly API for programmatic use (v0.10.0, March 2026)
The tag management feature makes this server particularly useful for PKM workflows that rely on tagging taxonomies for organization.
obsidian-mcp (StevenStavrakis)
Repository: StevenStavrakis/obsidian-mcp | Stars: ~720 | Type: Community | Language: TypeScript
A simpler, lightweight Obsidian MCP server focused on clean, reliable vault operations without the complexity of full-featured alternatives.
mcpvault (bitbonsai)
Repository: bitbonsai/mcpvault | Stars: ~1,600 | Type: Community | Language: TypeScript
A lightweight Obsidian MCP server for safe vault access. Works with Claude, ChatGPT, and any MCP client; 200+ commits with active maintenance as of mid-2026. Does not require the Local REST API plugin for some operations, making setup simpler for users who prefer not to install extra Obsidian plugins.
obsidian-mcp-plugin (aaronsb)
Repository: aaronsb/obsidian-mcp-plugin | Stars: ~440 | Type: Community | Language: TypeScript
Runs as an Obsidian plugin itself — the MCP server process runs inside Obsidian via HTTP transport rather than as an external Node.js process. Supports semantic vault operations; last released July 2026.
Obsidian Vault as MCP (ebullient)
Repository: ebullient/obsidian-vault-mcp | Type: Community plugin | Plugin downloads: ~5,000
An Obsidian community plugin (installable from the Obsidian plugin marketplace) that runs an MCP server inside Obsidian. Works with Open WebUI and Claude Desktop. A distinct category from external MCP servers that call Obsidian’s REST API — the server runs inside the app rather than alongside it.
Obsidian MCP Ecosystem Notes
The Obsidian MCP ecosystem is notably fragmented — more than 100 entries appear in MCP registries like PulseMCP, including many forks and variants. Roughly a dozen meaningfully distinct implementations exist. Differentiation comes in:
- Frontmatter handling — some servers understand YAML frontmatter as structured data, others treat it as plain text
- Link awareness — some servers can follow
[[wikilinks]]and resolve backlinks, others cannot - Search sophistication — ranging from simple string matching to semantic search with embeddings
- Write safety — some servers are read-only by default, others allow full modification
- Plugin requirement — most require the Local REST API plugin; some (mcpvault, StevenStavrakis) read vault files directly from disk
When choosing an Obsidian MCP server, the critical question is whether you need read-only access (for research and retrieval) or read-write access (for automated note creation and modification).
Notion MCP Servers
Notion’s MCP ecosystem is the most officially supported of any PKM tool. Notion has invested in both a hosted MCP server (no self-hosting required) and an open-source server, plus the community has built more than 50 additional implementations according to PulseMCP.
Official Notion MCP Server (makenotion)
Repository: makenotion/notion-mcp-server | Type: Official | Language: TypeScript
The official Notion MCP server wraps the Notion API and exposes it through the Model Context Protocol. It’s available in two deployment modes.
Key capabilities:
- search — find pages and databases across your workspace
- retrieve_page — get full page content including blocks and properties
- query_database — filter, sort, and paginate through database entries
- create_page — build new pages with rich content blocks
- update_page — modify page properties and content
- Hosted deployment — Notion runs the MCP server for you; no infrastructure needed
- Self-hosted deployment — run your own instance with the open-source server
The hosted MCP server is optimized for token efficiency, returning Markdown-based content rather than raw API JSON. This makes it significantly cheaper and more effective when used with LLMs. Version 2.0.0 migrated to the Notion API 2025-09-03, introducing data sources as the primary abstraction for databases.
mcp-notion-server (suekou)
Repository: suekou/mcp-notion-server | Stars: ~914 | Type: Community | Language: TypeScript
The most popular community Notion MCP server. Targets the Notion API 2026-03-11 version and is designed for compact, AI-friendly responses that reduce token usage compared to raw Notion API JSON.
Notion MCP via Composio
Platform: Composio (mcp.composio.dev/notion) | Type: Integration platform
Composio wraps the Notion API in an MCP-compatible interface as part of its broader AI agent tool platform, adding managed authentication and additional tooling.
Notion Ecosystem Notes
Notion’s official MCP server is the gold standard for how PKM vendors should approach AI integration:
- Hosted option eliminates infrastructure burden — users don’t need to run anything
- Token-optimized output — Markdown instead of JSON reduces costs
- Official support and updates — tied to Notion’s API versioning
- Community alternatives — more than 50 Notion-related MCP entries on PulseMCP for specialized use cases
The main limitation is that Notion is cloud-first — all data goes through Notion’s servers, unlike local-first tools like Obsidian where data stays on your machine.
Roam Research MCP Server
roam-research-mcp (2b3pro)
Repository: 2b3pro/roam-research-mcp | Type: Community | Language: TypeScript
Roam Research’s outliner-based, bidirectional-linking approach to note-taking gets MCP access through this community server. It provides comprehensive API access to Roam Research graphs.
Key capabilities:
- Graph traversal — navigate the block-based structure and follow bidirectional links
- Block operations — read, create, and modify individual blocks within pages
- Search — find content across the graph using Roam’s query syntax
- Page management — create and modify pages with their block hierarchies
- Advanced data retrieval — access the graph’s link structure programmatically
Roam’s block-based structure and bidirectional links make it particularly well-suited to MCP-based knowledge retrieval — the graph structure gives AI agents a natural way to traverse related concepts rather than searching linearly.
Logseq MCP Servers
mcp-logseq (ergut)
Repository: ergut/mcp-logseq | Stars: ~313 | Type: Community | Language: Python
The most popular Logseq MCP server. Supports semantic vector search across notes and is compatible with Logseq’s DB-mode graphs (Logseq’s newer SQLite-based storage format).
Key capabilities:
- Semantic vector search — query notes by meaning, not just keywords
- DB-mode graph support — works with both the older file-based and newer SQLite-based Logseq formats
- Page and block operations — read and create notes within the Logseq graph
mcp-pkm-logseq (ruliana)
Repository: ruliana/mcp-pkm-logseq | Type: Community | Language: Python
An earlier Logseq MCP server with customizable AI instructions. Uses Logseq’s HTTP API and filters notes by topic and date, allowing focused retrieval from large knowledge bases.
Key capabilities:
- Customizable AI instructions — configure how the AI agent interacts with your knowledge base
- Advanced search — query across pages, blocks, and properties
- Content creation — create and modify pages and blocks within the Logseq graph
- Knowledge management — leverage Logseq’s outliner structure and bidirectional links
Logseq is open-source and local-first (like Obsidian), which means MCP servers for Logseq work with local files — no cloud API required.
Evernote MCP Servers
Despite being one of the oldest note-taking platforms, Evernote has a modest but functional MCP ecosystem.
mcp-evernote (verygoodplugins)
Repository: verygoodplugins/mcp-evernote | Type: Community | Language: TypeScript
A full-featured Evernote MCP server with write capabilities and proper authentication.
Key capabilities:
- OAuth 1.0a authentication — secure access to Evernote accounts
- Full CRUD operations — create, read, update, and delete notes
- Notebook management — browse and organize notebooks
- Tag operations — manage Evernote’s tagging system
- Note synchronization — keep AI agent actions in sync with Evernote’s sync system
evernote-mcp-server (brentmid)
Repository: brentmid/evernote-mcp-server | Type: Community | Language: TypeScript
A read-only Evernote MCP server focused on querying and retrieving notes. Useful for users who want AI access to their Evernote archive without the risk of accidental modification.
Key capabilities:
- Search and query — find notes by keyword, notebook, or tag
- Read notes — retrieve full note content
- List notebooks and tags — browse the organizational structure
- Read-only by design — no write operations prevent accidental changes
Apple Notes MCP Servers
Platform: macOS only | Type: Community
Multiple MCP servers exist for Apple Notes on macOS. Apple Notes has no REST API; all servers access it through AppleScript or the Notes SQLite database.
Notable implementations:
- RafalWilinski/mcp-apple-notes (~407 stars) — TypeScript; on-device embeddings for semantic search and RAG; no external API keys required
- bfeeny/apple-notes-mcp — AppleScript bridge with full CRUD operations
- Dan8Oren/mcp-apple-notes — semantic search + folder browsing + RAG, fully local
All implementations are macOS-only. The on-device embedding approach (RafalWilinski’s) is notable for keeping all data local — no cloud API required for semantic search.
Reference Management: Zotero MCP
zotero-mcp (54yyyu) — recommended
Repository: 54yyyu/zotero-mcp | Stars: ~4,400 | Type: Community | Language: Python
The most popular Zotero MCP server. Connects to Zotero’s local HTTP API and provides semantic search, annotation extraction, and citation analysis.
Key capabilities:
- Semantic search — find papers by meaning, not just keywords
- Annotation extraction — access highlights and notes from PDFs in your library
- Citation analysis — understand relationships between papers
- Library browsing — navigate collections, items, and attachments
zotero-mcp (kujenga)
Repository: kujenga/zotero-mcp | Stars: ~158 | Type: Community | Language: Python
An alternative that uses the Zotero web API (rather than the local HTTP API), useful for accessing Zotero libraries without running the desktop app.
For researchers and academics, Zotero MCP is one of the highest-value PKM integrations — it lets AI agents access your entire research library, understand citation relationships, and help synthesize findings across papers.
Read-Later and Bookmarking
Readwise MCP Server (Official)
Documentation: docs.readwise.io/tools/mcp | Type: Official | Endpoint: mcp2.readwise.io/mcp
Readwise’s official MCP server provides access to your reading highlights and documents. Readwise is widely used in the PKM community as a bridge between reading and note-taking.
Key capabilities (16 tools total):
- Document search — hybrid vector + full-text search across books, articles, and other sources
- Highlight retrieval — access highlighted passages with their context; create, update, and delete highlights
- Spaced repetition — access your daily review queue
- Readwise Reader integration — save new URLs or text content, move documents between locations (inbox, archive, shortlist), bulk edit up to 50 documents at once
- Tag management — manage tags across your reading library
- Authentication — OAuth-based (browser redirect to Readwise account)
The combination of Readwise MCP with an Obsidian or Notion MCP server creates a powerful reading-to-notes pipeline: AI agents can pull relevant highlights from what you’ve read and incorporate them into your notes.
Raindrop.io MCP Server (Official)
Documentation: developer.raindrop.io/mcp | Type: Official | Status: Beta | Tier: Pro subscribers only | Also: help.raindrop.io/mcp
Raindrop.io provides an official MCP server for its bookmark management platform, currently in beta and restricted to Pro subscribers. The server endpoint is https://api.raindrop.io/rest/v2/ai/mcp with OAuth 2.1 authorization.
Key capabilities:
- Bookmark search — find saved bookmarks by content, tag, or collection
- Collection management — browse and organize bookmark collections
- Bookmark modification — add, update, and delete bookmarks
- Tag and highlight management — manage tags and highlights within bookmarks
- Supported clients — Claude, ChatGPT, VS Code/GitHub Copilot, Cursor, Windsurf, Zed, and others
For knowledge workers who use Raindrop.io as their web clipping tool, this server turns a passive bookmark archive into an active knowledge source. Note the Pro subscription requirement and current beta status.
Pocket MCP Server
Repository: kazuph/mcp-pocket | Type: Community | Language: TypeScript
An MCP server for the Pocket read-later service, providing pocket_get_articles (fetch saved articles with title, URL, and excerpt) and pocket_mark_as_read (archive by item ID). Requires a Pocket Consumer Key and Access Token from the Pocket Developer Portal.
Team Knowledge Bases
Slite MCP (Official)
Landing page: slite.com/mcp | Type: Official | Endpoint: api.slite.com/mcp
Slite is a team knowledge management platform, and its official MCP server exposes 40+ tools for working with the Slite workspace. No GitHub repo or local installation required — the server runs as a remote hosted endpoint.
Key capabilities:
- Document operations — search, create, update, archive, and restore documents
- Collection management — browse and organize Slite collections
- Block-level editing — fine-grained content modifications at the block level
- Comment access — retrieve and work with document comments
- Parallel queries — run multiple search queries simultaneously
For teams already using Slite as their central knowledge base, this MCP server brings AI agents directly into that workflow without requiring any local setup.
More Note-Taking Platforms
Anytype MCP (Official)
Repository: anyproto/anytype-mcp | Stars: ~484 | Type: Official | Language: TypeScript | Package: @anyproto/anytype-mcp
Anytype is a local-first, privacy-focused PKM tool with its own object graph model. Its official MCP server (published by the anyproto organization, Anytype’s development team) connects to the Anytype desktop app running locally at http://127.0.0.1:31009. Last updated June 2026 (v1.2.9).
Key capabilities:
- Space search — search across all Anytype spaces
- Object management — create and manage objects (Anytype’s equivalent of notes)
- Properties, tags, types, templates — interact with Anytype’s structured data model
Bear MCP Servers (Community)
Notable repos: bejaminjones/bear-notes-mcp, netologist/mcp-bear-notes | Type: Community | Platform: macOS only
Bear is a popular Markdown-based note-taking app for macOS and iOS. It has no official REST API, but Bear stores its notes in a local SQLite database that community MCP servers access directly.
Notable implementations:
- bejaminjones/bear-notes-mcp — the most complete implementation, supporting full read and write operations including create, update, archive, and tag management
- netologist/mcp-bear-notes — focused on search and retrieval
All implementations are macOS-only. The SQLite-based approach means no Bear API key or plugin is required.
Craft MCP (Official)
Documentation: craft.do/imagine/guide/mcp/mcp | Type: Official | Endpoint: mcp.craft.do
Craft is a document editor for macOS and iOS. Its official MCP server runs as a remote hosted endpoint with OAuth authorization — no local installation required.
Key capabilities (32+ tools):
- Document operations — read, summarize, update, and search documents across spaces
- Tag and date filtering — filter searches by tag or date range
- Daily notes — access and update your daily note
- Task management — create and update tasks within documents
- Collection management — browse and organize document collections
Day One MCP (Official)
Repository: bloom/dayone-mcp-server | Type: Official | Language: TypeScript
Day One is a journaling app. Its official MCP server (from the bloom organization, Day One’s development team) runs as a local stdio server and requires enabling journal access in Day One’s preferences.
Key capabilities:
- Journal entries — create and update entries with Markdown content
- Full-text search — search across all journals with date-range filtering
- Tag management — manage tags on journal entries
- Date-range retrieval — retrieve entries from specific time periods
Amplenote MCP (Official)
Documentation: amplenote.com/help/ai_mcp_server_connect_amplenote_claude_codex | Type: Official | Tier: Unlimited plan required
Amplenote integrates an MCP server directly into its desktop app — no separate installation or GitHub repo. The server runs locally at http://127.0.0.1:39377/mcp.
Key capabilities:
- Note operations — read, search, create, and update notes
- Task management — create and update tasks
- Workspace export — expose the full workspace as Markdown
The built-in approach means zero setup friction for existing Amplenote users on the Unlimited plan.
RemNote MCP (Community)
Repository: robert7/remnote-mcp-server | Type: Community
RemNote is a note-taking app that combines spaced-repetition flashcards with a hierarchical note structure. The community MCP server connects through the RemNote Automation Bridge plugin (requires installation in the RemNote app).
Key capabilities:
- Note and flashcard creation — create notes, flashcards, and hierarchical trees
- Search — full-text and tag-based search
- Daily journal — append to daily journal entries
- Content modification — update note content and tags
Capacities MCP (Community)
Repository: inconceivablelabs/capacitiesMCP | Type: Community
Capacities is a structured PKM tool that organizes knowledge into typed objects (notes, people, books, etc.). No official MCP server exists; the community implementation uses the Capacities REST API v2 for full CRUD operations on objects.
Whiteboard and Visual PKM
Heptabase MCP Server (Official)
Documentation: support.heptabase.com | Type: Official
Heptabase is a visual knowledge management tool that combines note-taking with whiteboard-style spatial organization. Its official MCP server lets AI agents interact directly with your Heptabase space.
Key capabilities:
- Read notes and whiteboards — AI agents can discover and read existing cards and whiteboard layouts
- Understand context — agents see how cards relate spatially on whiteboards
- Search knowledge base — find relevant cards and content across your space
- Write back — save new insights as note cards or journal entries in your Heptabase space
- Cross-tool integration — works with ChatGPT, Claude, and other MCP-compatible tools
Heptabase’s spatial approach to knowledge management is uniquely suited to MCP — the whiteboard structure gives AI agents a visual context for understanding how ideas relate, not just their textual content.
Tana MCP Server
Type: Official API + MCP
Tana released a full API and introduced an MCP server that streamlines the process of connecting AI agents to Tana’s structured note-taking system.
Key capabilities:
- Node operations — access Tana’s node-based structure programmatically
- Tana Tracker Viz — MCP-triggered visualization that generates GitHub-style contribution graphs from Tana nodes
- Habit and project tracking — interact with tracked nodes for habits, supplements, and projects
- Structured data — leverage Tana’s supertag and field system through MCP
Wiki-Style and Self-Hosted Notes
Joplin MCP Servers
Repositories: dweigend/joplin-mcp-server, alondmnt/joplin-mcp | Type: Community
Joplin is an open-source, privacy-focused note-taking application. Multiple MCP server implementations exist.
Key capabilities:
- Note access — read and search notes across notebooks
- Notebook management — navigate Joplin’s notebook hierarchy
- Tag operations — manage tags and tag associations
- Python and TypeScript options — dweigend’s server uses TypeScript; alondmnt’s uses Python (FastMCP) with the joppy library
- End-to-end encryption compatible — works with Joplin’s encryption when properly configured
Trilium/TriliumNext MCP Servers
Repositories: aimbitgmbh/trillium-mcp, tan-yong-sheng/triliumnext-mcp, pwelty/mcp_trilium, mursilsayed/trilium-bolt | Type: Community
Trilium Notes (and its community fork TriliumNext) is a hierarchical note-taking application popular for personal wikis and knowledge bases. Multiple MCP servers provide access.
Key capabilities:
- ETAPI access — connect through Trilium’s External API
- Markdown-friendly output — trilium-bolt returns LLM-friendly markdown instead of raw HTML
- Search, create, update — full note management operations
- Hierarchical navigation — traverse Trilium’s tree structure
- Natural language interaction — search and manage notes conversationally through Claude
The four competing Trilium MCP servers reflect the tool’s popularity among self-hosting enthusiasts who want AI access to their personal wikis.
SilverBullet MCP Server
Repository: Ahmad-A0/silverbullet-mcp | Type: Community
SilverBullet is an open-source, self-hosted, wiki-style note-taking application. Its MCP server exposes notes and data through standardized MCP tools and resources.
Knowledge Graph Memory Servers
Beyond specific note-taking apps, a category of MCP servers provides persistent memory through knowledge graph structures — effectively creating a “second brain” that exists purely within the MCP ecosystem.
Knowledge Graph Memory Server (Anthropic Official)
Repository: modelcontextprotocol/servers (see src/memory) | Type: Official (Anthropic) | Monorepo stars: ~88,700
The most widely deployed memory MCP server, maintained as part of Anthropic’s official MCP server collection. Note: modelcontextprotocol/servers is a monorepo containing many reference server implementations — the ~88,700 stars reflect the whole collection, not the memory server specifically (there is no standalone memory-server repository). It enables persistent memory for Claude through a local knowledge graph stored as JSON.
Key capabilities:
- Entity management — create, read, update entities with properties
- Relationship tracking — define typed relationships between entities
- Persistent storage — knowledge persists between conversations in a local file
- Simple architecture — no external database required; stores everything in a JSON file
This server is frequently used as a PKM layer by users who don’t have an existing note-taking tool — it provides basic “remember this” functionality that persists across Claude conversations.
Cognee
Repository: topoteretes/cognee | Stars: ~29,000 | Type: Community/Commercial
A memory manager for AI apps and agents that supports various graph and vector stores, allowing ingestion from 30+ data sources. More suitable for developer and enterprise use cases than personal PKM.
Basic Memory (basicmachines-co)
Repository: basicmachines-co/basic-memory | Stars: ~3,500 | Type: Community
A local-first knowledge management system that builds a semantic graph from Markdown files. All knowledge is stored in standard Markdown files on your computer.
Key capabilities:
- Markdown-native — all knowledge stored as Markdown files you can open in any editor
- Semantic graph — automatically builds a knowledge graph from file links and content
- Obsidian compatible — Markdown files work seamlessly in Obsidian
- No cloud dependency — everything stays local
- Conversation persistence — AI conversations become searchable knowledge
Basic Memory addresses a fundamental PKM problem: conversations with AI assistants are ephemeral. By storing conversation insights as Markdown files with semantic connections, it turns every AI interaction into a building block for your knowledge base.
mem0
Type: Commercial | Cloud MCP: app.mem0.ai
Provides MCP-compatible knowledge graph memory for enterprise AI deployments, focusing on persistent memory across agent pipelines (LangGraph, CrewAI, AutoGen). The self-hosted open-source repo (mem0ai/mem0-mcp) was archived in March 2026; the active path is the cloud-hosted MCP endpoint at app.mem0.ai.
PKM MCP Server Comparison
| Platform | Official MCP | Community Servers | Local-First | Write Support | Auth Method |
|---|---|---|---|---|---|
| Obsidian | No | 100+ (incl. forks) | Yes | Yes | Local REST API |
| Notion | Yes (hosted + self-hosted) | 50+ | No (cloud) | Yes | OAuth/API key |
| Roam Research | No | 1+ | No (cloud) | Yes | API key |
| Logseq | No | 2+ | Yes | Yes | Local files |
| Evernote | No | 2+ | No (cloud) | Yes/Read-only | OAuth 1.0a |
| Apple Notes | No | 3+ | Yes (macOS) | Yes | Local (AppleScript) |
| Zotero | No | 2+ | Yes | Limited | Local API |
| Readwise | Yes | 1+ | No (cloud) | Yes | OAuth |
| Raindrop.io | Yes (beta, Pro) | 1+ | No (cloud) | Yes | OAuth 2.1 |
| Heptabase | Yes | — | No (cloud) | Yes | API key |
| Tana | Yes (via API) | — | No (cloud) | Yes | API key |
| Joplin | No | 2+ | Yes | Yes | Local API |
| Trilium | No | 4+ | Yes | Yes | ETAPI |
| SilverBullet | No | 1 | Yes | Yes | Local |
| Basic Memory | N/A (is MCP) | — | Yes | Yes | Local files |
| Anytype | Yes | 1+ | Yes | Yes | Local HTTP API |
| Bear | No | 3+ | Yes (macOS) | Yes | SQLite (local) |
| Craft | Yes (hosted) | — | No (cloud) | Yes | OAuth |
| Day One | Yes | 2+ | Yes (local) | Yes | Local (stdio) |
| Amplenote | Yes (built-in) | — | No (cloud) | Yes | Local HTTP |
| Slite | Yes (hosted) | — | No (cloud) | Yes | OAuth |
| RemNote | No | 1+ | No (cloud) | Yes | Plugin API |
| Capacities | No | 2+ | No (cloud) | Yes | REST API |
Architecture Patterns
Pattern 1: AI-Augmented Second Brain
┌─────────────────────────────────────────────┐
│ AI Assistant (Claude) │
│ │
│ "What do I know about distributed systems?" │
│ "Summarize my notes from last week" │
│ "Find connections between these topics" │
└──────┬──────────────┬──────────────┬─────────┘
│ │ │
▼ ▼ ▼
┌──────────┐ ┌──────────────┐ ┌──────────┐
│ Obsidian │ │ Readwise │ │ Zotero │
│ MCP │ │ MCP │ │ MCP │
│ Server │ │ Server │ │ Server │
└──────┬───┘ └──────┬───────┘ └──────┬───┘
│ │ │
▼ ▼ ▼
┌──────────┐ ┌──────────────┐ ┌──────────┐
│ Obsidian │ │ Reading │ │ Research │
│ Vault │ │ Highlights │ │ Papers │
│ (local) │ │ (cloud) │ │ (local) │
└──────────┘ └──────────────┘ └──────────┘
This pattern connects multiple knowledge sources through MCP, giving the AI agent a comprehensive view of your knowledge base. The agent can correlate highlights from your reading with notes in your vault and references in your Zotero library.
Pattern 2: Automated Knowledge Capture Pipeline
┌──────────┐ ┌──────────┐ ┌──────────────┐
│ Web │ │ Books/ │ │ Research │
│ Reading │ │ Articles│ │ Papers │
└────┬─────┘ └────┬─────┘ └──────┬───────┘
│ │ │
▼ ▼ ▼
┌──────────┐ ┌──────────┐ ┌──────────────┐
│Raindrop │ │ Readwise │ │ Zotero │
│ MCP │ │ MCP │ │ MCP │
└────┬─────┘ └────┬─────┘ └──────┬───────┘
│ │ │
└───────────────┼────────────────┘
│
▼
┌─────────────────────┐
│ AI Agent (Claude) │
│ │
│ • Extract insights │
│ • Create backlinks │
│ • Add metadata │
│ • Generate summaries│
└──────────┬──────────┘
│
▼
┌─────────────────────┐
│ Obsidian/Notion │
│ MCP Server │
│ │
│ → Literature notes │
│ → Concept notes │
│ → Index updates │
└─────────────────────┘
This pipeline automates the “progressive summarization” workflow popularized by Tiago Forte: content flows from capture tools through AI processing into structured notes in your primary PKM tool.
Pattern 3: Cross-Platform Knowledge Sync
┌─────────────────────────────────────────┐
│ AI Agent (Claude) │
│ │
│ "Sync my Notion project notes to │
│ Obsidian with proper backlinks" │
└──────────┬───────────────┬──────────────┘
│ │
▼ ▼
┌──────────┐ ┌──────────┐
│ Notion │ │ Obsidian │
│ MCP │◄──►│ MCP │
│ Server │ │ Server │
└──────┬───┘ └──────┬───┘
│ │
▼ ▼
┌──────────┐ ┌──────────┐
│ Notion │ │ Obsidian │
│Workspace │ │ Vault │
│ (cloud) │ │ (local) │
└──────────┘ └──────────┘
Many knowledge workers use multiple PKM tools — perhaps Notion for collaborative work and Obsidian for personal notes. MCP enables AI agents to bridge these tools, reading from one and writing to another with appropriate format conversion.
Pattern 4: Persistent Memory Layer
┌─────────────────────────────────────────┐
│ AI Assistant Conversations │
│ │
│ Conversation 1 → Conversation 2 → ... │
└──────────────────┬──────────────────────┘
│
▼
┌─────────────────────┐
│ Knowledge Graph │
│ Memory MCP Server │
│ │
│ Entities: │
│ • Projects │
│ • People │
│ • Decisions │
│ • Preferences │
└──────────┬──────────┘
│
▼
┌─────────────────────┐
│ Local JSON/MD │
│ Knowledge Store │
│ │
│ ┌───────────────┐ │
│ │ Entity Graph │ │
│ │ with Relations│ │
│ └───────────────┘ │
└─────────────────────┘
This pattern uses knowledge graph memory servers to give AI assistants persistent memory across conversations — effectively creating a PKM system that exists purely within the AI interaction layer. Basic Memory extends this by storing the graph as readable Markdown files.
Ecosystem Gaps and Opportunities
Despite the density of the PKM MCP ecosystem, significant gaps remain:
No unified PKM protocol. Each MCP server implements its own tool set with different naming conventions, capabilities, and data models. There’s no standard set of PKM operations (create note, search, link, tag) that all servers implement consistently.
Limited graph awareness. Most Obsidian and Notion MCP servers treat notes as isolated documents. Few can traverse backlinks, understand graph structure, or reason about relationships between notes — which is the core value proposition of modern PKM tools.
No mainstream migration tools. MCP could enable seamless migration between PKM tools (read from Evernote, write to Obsidian with format conversion), but no MCP server currently focuses on this use case.
Missing or early-stage platforms. Several popular PKM and note-taking tools lack maintained MCP servers:
- Google Keep — feuerdev/keep-mcp (~85 stars) exists as a community server using the reverse-engineered gkeepapi library, but no official server
- Supernotes — the team announced OAuth 2.1 work in June 2026 but no official server has shipped yet
- Standard Notes — no MCP server found as of July 2026
- Reflect — no MCP server found
- DEVONthink — no MCP server for this powerful macOS knowledge management tool
- Dendron — no MCP server (project was discontinued in 2023)
No AI writing assistants with PKM context. While MCP servers give AI agents access to your knowledge base, no server specifically focuses on the writing workflow — drafting with awareness of your existing notes, voice, and style conventions.
Limited mobile support. Most PKM MCP servers require a running desktop application or local server. Mobile-first workflows (capture on phone, process on desktop) have no MCP bridge.
Getting Started
For Obsidian Users
- Install the Obsidian Local REST API community plugin
- Choose an MCP server — start with mcp-obsidian (MarkusPfundstein) for simplicity or obsidian-mcp-server (cyanheads) for tag management
- Configure the server with your vault path and REST API key
- Start with read-only queries before enabling write operations
For Notion Users
- Use Notion’s official hosted MCP server for the simplest setup — no infrastructure required
- Connect through Claude Desktop, VS Code, or ChatGPT
- Start with search and retrieve operations to understand what data is accessible
- For advanced use cases, self-host the open-source server for customization
For Researchers
- Set up 54yyyu/zotero-mcp for reference library access
- Add Readwise MCP (docs.readwise.io/tools/mcp) for reading highlights
- Connect your primary note-taking tool (Obsidian, Notion, or Logseq)
- Use the AI agent to create literature notes and synthesize findings across sources
For PKM Beginners
- Start with Anthropic’s Knowledge Graph Memory server — it requires no existing PKM tool
- Use it to build persistent memory across AI conversations
- Consider Basic Memory if you want your knowledge stored as readable Markdown files
- Graduate to a full PKM tool (Obsidian, Notion, or Logseq) when your knowledge base grows
For Self-Hosters
- Joplin and Trilium/TriliumNext offer fully self-hosted note-taking with MCP access
- SilverBullet provides wiki-style self-hosted notes with MCP
- All local-first tools (Obsidian, Logseq, Joplin, Trilium, Anytype) keep your data on your own machine
- Combine with Basic Memory for AI conversation persistence without cloud dependencies
For Multi-Tool Users
- Identify your “source of truth” PKM tool and set up its MCP server first
- Add MCP servers for capture tools (Readwise, Raindrop.io, Pocket)
- Use AI agents to automate the flow from capture to structured notes
- Consider the cross-platform sync pattern for bridging collaborative and personal tools
For macOS Note-Taking Apps
Craft, Bear, Day One, and Apple Notes all have MCP servers (official or community). Bear and Apple Notes use local storage; Craft (mcp.craft.do) and Day One (bloom/dayone-mcp-server) run as hosted or local stdio servers. All are macOS-only or macOS-primary.
Conclusion
The PKM MCP ecosystem is one of the richest in the entire MCP landscape, with MCP servers across note-taking platforms, reference managers, bookmark tools, journaling apps, team knowledge bases, and knowledge graph memory systems. The density reflects a fundamental truth: personal knowledge bases are exactly the kind of rich, structured context that makes AI agents dramatically more useful.
The most impactful PKM MCP pattern isn’t any single tool integration — it’s the ability to connect multiple knowledge sources through a single AI agent. An agent that can simultaneously access your Obsidian vault, Readwise highlights, Zotero references, and Raindrop bookmarks can synthesize knowledge in ways that no single tool can achieve alone.
The ecosystem is still maturing. Graph awareness, cross-tool standards, and mobile support remain gaps. But for knowledge workers willing to set up MCP servers for their tools, the payoff is significant: your second brain becomes a collaborative partner rather than a passive archive.
This guide was researched and written by AI as part of ChatForest, a project by Rob Nugen. We do not claim hands-on testing of any servers mentioned — our analysis is based on published documentation, open-source repositories, and community discussions. Last updated: July 22, 2026.