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

Anytype MCP Server

Official
Official MCP server that lets AI assistants manage your encrypted Anytype wiki in natural language.
How to install ↓ Compare GitHub source repository ↗
Category
Collaboration #2 of 55
Stars
★ 537 Very popular
Transport
stdio (local process)
Runtime
Node.js · Bun
Credentials
API key / credential required
License
MIT
Last commit
64FMRS · C

Anytype MCP Server is the official Anytype-maintained MCP server, converting Anytype's OpenAPI specification into MCP tools so AI assistants can manage your encrypted, local and collaborative wiki in natural language. It runs over stdio via npm, is simple to configure, requires an API key and a local Anytype service, and suits existing Anytype users who want to bring AI into their workflow.

Strongest · Setup experience 15/20 Weakest · Reliability 10/20

Reliability
10/20
Security and permissions
11/20
Maintenance
14/20
Documentation
14/20
Setup experience
15/20
Why each score
Reliability 10/20
The only supplied material is the README, the server.json manifest, and repository metadata; no source files, package.json, tests, or CI workflows are provided. We can see the npm identifier, version 1.0.7, and a declared stdio transport, but there are no committed tests and no CI evidence that the MCP init and tool-list handshake completes, so the static calibration caps reliability at 12. The server claims to auto-generate MCP tools from Anytype's OpenAPI spec (global/space search, spaces and members, objects and lists, properties and tags, types and templates); without the generation logic and error-mapping code, argument validation, failure diagnostics, and dependency behavior are uncontrollable and unverifiable. 26 open issues further suggest edge paths may be rough. Score 10: the happy path is plausible, but tests, error handling, and dependency boundaries are opaque.
Security and permissions 11/20
Credential handling: OPENAPI_MCP_HEADERS is marked is_secret and required, and the README instructs users to embed a Bearer API key as a JSON string directly in client config files, a common but weak practice with no OS keychain or encrypted storage support. On the plus side, the key is locally generated by the Anytype app and the server defaults to a loopback endpoint (http://127.0.0.1:31009), so the data boundary is largely local and external network exposure is limited. No malware, covert exfiltration, fabricated maintainer identity, or example containing real tokens was found, so no red line applies. Deductions: no disclosed confirmation or dry-run mechanism for dangerous operations (deleting objects/spaces, member changes, bulk edits), no least-privilege or scope statement, and ANYTYPE_API_BASE_URL can redirect traffic to an arbitrary host without any accompanying warning. Score 11: main risks are visible and no red line, but permission scoping and confirmation are incomplete.
Maintenance 14/20
There is clear organizational ownership (Anytype/anyproto, a Swiss association) and an MIT license; the README provides a security contact ([email protected]) plus links to code-of-conduct and contributing guides; the manifest shows an active 1.0.7 release and ongoing npm publishing, so the maintenance chain is broadly sound. Deductions: no commit cadence, release timeline, dependency-update policy, or support window is given, and there is no evidence of issue response for the 26 open issues; versioning and governance details must be inferred. Score 14: license, ownership, and security channel are clear, but versioning and governance transparency have gaps.
Documentation 14/20
Documentation is layered: installation (npx, global install, claude mcp add), two ways to obtain an API key (in-app and CLI), config snippets for several clients (Claude Desktop, Cursor, Windsurf, Raycast, LM Studio, Kiro), custom API base URL with port notes (31009/31012), and natural-language interaction examples that are copy-paste ready. Deductions: there is no tool inventory or per-tool parameter/return documentation, no pagination, rate-limit, cost, or troubleshooting section, and the exact set and naming of the OpenAPI-derived tools is undisclosed, so users cannot predict which capabilities the assistant will call; error formats are also undocumented. Score 14: usable to get started, but with hidden assumptions and troubleshooting gaps.
Setup experience 15/20
The install path is short and clear: `npx -y @anyproto/anytype-mcp` runs the server, and the docs supply mainstream client JSON configs plus a single claude mcp add command and one-click deeplinks, so a user only needs to substitute an API key to connect; no leaked tokens were observed. Deductions: it still depends on the Anytype desktop app enabling its local API and issuing a key, and the differing default local port versus anytype-cli's port requires manual ANYTYPE_API_BASE_URL configuration; no Node/Bun runtime version requirements, Windows/macOS/Linux compatibility notes, or troubleshooting steps are given (the from-source build requires Bun, an extra barrier for newcomers). Under the static calibration, without real CI and test evidence setup may not exceed 15. Score 15: solid config examples and a clear path, minus external dependencies and absent automated verification.

Static review · not runListed 2026-09-18

Read the FMRS scoring method →

Fit and risk

What it can accessUses the networkConnects to a database

Best for

  • Users who already rely on Anytype as a personal or collaborative knowledge base
  • Users who want to create and manage Anytype objects quickly with natural language
  • Advanced users running the Anytype desktop app locally who want AI assistant integration

Not for

  • Users without Anytype installed or running and unable to obtain an API key
  • Scenarios needing a remotely hosted service instead of a local stdio process
  • Users who want zero configuration without setting environment variables

Required permissions

  • Requires an Anytype API key passed via an Authorization: Bearer header
  • Needs access to the local Anytype API service (default http://127.0.0.1:31009)
  • Passes authentication headers through the OPENAPI_MCP_HEADERS environment variable, treated as secret
  • Runs an npx command locally as a stdio subprocess

Risks and side effects

  • OPENAPI_MCP_HEADERS contains the API key; leaking this configuration can expose Anytype data
  • The server can create and modify spaces, objects and lists on behalf of the key owner
  • Running npx -y downloads and executes package code from npm
  • Pointing ANYTYPE_API_BASE_URL at an untrusted address can send authentication headers there

Setup

Before you start

Runtime：Node.js · Bun

OPENAPI_MCP_HEADERS requiredsecret A JSON string with Authorization Bearer <YOUR_API_KEY> and Anytype-Version: 2025-11-08; create the API key in Anytype under App Settings > API Keys.
Other optional settings (1)
ANYTYPE_API_BASE_URL optional Optional; defaults to http://127.0.0.1:31009. Set it for anytype-cli (port 31012) or another custom base URL.
  1. Get your API key: open Anytype, go to App Settings, navigate to the API Keys section and click Create new. Alternatively run: npx -y @anyproto/anytype-mcp get-key.
  2. Add the following to your MCP client (Claude Desktop, Cursor, Windsurf, Raycast, etc.), replacing <YOUR_API_KEY>: {"mcpServers":{"anytype":{"command":"npx","args":["-y","@anyproto/anytype-mcp"],"env":{"OPENAPI_MCP_HEADERS":"{\"Authorization\":\"Bearer <YOUR_API_KEY>\", \"Anytype-Version\":\"2025-11-08\"}"}}}}.
  3. For Claude Code (CLI) run: claude mcp add anytype -e OPENAPI_MCP_HEADERS='{"Authorization":"Bearer <YOUR_API_KEY>", "Anytype-Version":"2025-11-08"}' -s user -- npx -y @anyproto/anytype-mcp.
  4. For a custom API base URL (e.g. anytype-cli on port 31012), set ANYTYPE_API_BASE_URL. You can also install globally with npm install -g @anyproto/anytype-mcp and start it with the command anytype-mcp.
claude_desktop_config.json 
{"mcpServers":{"anytype":{"command":"npx","args":["-y","@anyproto/anytype-mcp"],"env":{"OPENAPI_MCP_HEADERS":"{\"Authorization\":\"Bearer <YOUR_API_KEY>\", \"Anytype-Version\":\"2025-11-08\"}"}}}}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "anytype": {
      "command": "npx",
      "args": [
        "-y",
        "@anyproto/anytype-mcp"
      ],
      "env": {
        "OPENAPI_MCP_HEADERS": "{\"Authorization\":\"Bearer <YOUR_API_KEY>\", \"Anytype-Version\":\"2025-11-08\"}"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add anytype -e 'OPENAPI_MCP_HEADERS={"Authorization":"Bearer <YOUR_API_KEY>", "Anytype-Version":"2025-11-08"}' -- npx -y @anyproto/anytype-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After configuring OPENAPI_MCP_HEADERS in your MCP client, restart it and the anytype server's tools (search, spaces, objects, lists) should appear in the client's tool list; you can also run npx -y @anyproto/anytype-mcp get-key to confirm the package works and obtain an API key.

Troubleshooting

  1. Ensure the Anytype app is running and its API service is reachable (default 127.0.0.1:31009)
  2. Verify OPENAPI_MCP_HEADERS is valid JSON and the API key is correct
  3. If using anytype-cli (port 31012) or another address, confirm ANYTYPE_API_BASE_URL is set
  4. Check that the Anytype-Version header matches the target API version (example uses 2025-11-08)
  5. Copy the ready-to-use configuration snippet from the API Keys section in Anytype to avoid manual errors

Things to try

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

  • Create a new space called 'Project Ideas' with description 'A space for storing project ideas'
  • Add a new object of type 'Task' with title 'Research AI trends' to the 'Project Ideas' space
  • Create a collection called 'Tasks for this week' and add the two tasks to that list
  • Search across my spaces for objects related to LLMs

Use cases

Create a new space with a title and description, e.g. 'Project Ideas'
Add a new object such as a Task to a specific space
Set due dates on tasks and assign them to a member
Create a collection and add existing tasks to it
Manage objects, lists, properties, tags, types and templates conversationally

Supported clients

Claude Desktop
Cursor
Windsurf
Raycast
Claude Code
LM Studio
Kiro

Listed from the project's documentation, not tested by this site.

Overview

Anytype MCP Server is the official Model Context Protocol (MCP) server for Anytype. It converts Anytype's OpenAPI specification into MCP tools so AI assistants can interact with the Anytype API through natural language and manage your encrypted, local and collaborative wiki. Capabilities include global and space search, spaces and members, objects and lists, properties and tags, and types and templates. It runs over stdio and is distributed as the npm package @anyproto/anytype-mcp, requiring an OPENAPI_MCP_HEADERS environment variable containing Anytype API authentication headers. It connects to http://127.0.0.1:31009 by default, configurable via ANYTYPE_API_BASE_URL (for example port 31012 for anytype-cli).

Similar servers

Compare with other MCP servers

Source revision 8f50746e51e3 Data synced 2026-11-02 Read the FMRS scoring method