MeshWrite Documentation
Learn to organize writing nodes, connect AI agents with MCP, and use the MeshWrite Skill safely.
🌲 Introduction
Welcome to MeshWrite, the next-generation markdown editor engineered for creators who think in networks rather than lines. Traditional documents force a destructive edit cycle where overwriting drafts is the norm.
MeshWrite organizes writing as a tree of connected versions. Each project can contain branches that preserve alternate ideas and their relationship to the original.
Core Philosophy: Evolution is Creation
The canvas shows nodes and their branching relationships so you can compare different directions while retaining the context of how each version developed.
🔐 AI Providers and Data Flow
Configure model providers in Settings → AI Providers. MeshWrite's available provider and model choices can change over time.
KEY STORAGE AND REQUEST ROUTING
In the web app, provider keys are stored in your browser's localStorage. Cloud model requests are sent through the MeshWrite API proxy, which forwards the request and key to the selected provider. Review the data policies of MeshWrite and your model provider before sending sensitive content.
To add your credentials, open Settings → AI Providers, choose a provider, and configure its endpoint, model, and key.
Node IDs and project references
Every node has a stable, unique ID. The PWA shows a shortened form for scanning; copy the badge to get the full ID. Use this ID when an agent needs to read or refer to one exact node.
A reference such as #1.2 describes a node's position in one project's branch tree. It is useful to people, but can repeat in other projects. If using a reference with MCP, include its project ID; otherwise use the node's unique ID.
Workspace search finds nodes by title, content, ID, or project reference. The canvas also has a searchable node-list view for the current project.
Connect an AI client with MCP
MeshWrite provides a remote MCP server with OAuth. Add this server URL in an MCP-capable client, then sign in to MeshWrite and review the consent screen:
https://api.meshwrite.cc/mcp/oauth- Add the URL as a remote MCP server in your AI client.
- Continue through the MeshWrite sign-in page.
- Review and approve the requested read and write access.
- Return to the client and ask it to list your projects or search for a node.
OAuth is the recommended setup. A separate legacy MCP token connection remains available for clients that cannot use OAuth.
MCP tools and permissions
After authorization, the client can use the tools below. Access is limited to the signed-in account and to the operations these tools expose.
| Tool | What it does |
|---|---|
| list_projects | Lists your active projects and folders. |
| search_nodes | Finds nodes by text, hierarchical reference, or project. |
| get_node | Reads a node by unique ID, or by a project-scoped reference. |
| list_persons | Lists available AI Persons. |
| save_answer | Writes an answer as a private, unpublished draft in a chosen project. |
| run_person | Runs an AI Person on a node and saves the result as a branch or new version. |
Write access is not publishing access. MCP-created answers are drafts for you to review in MeshWrite. The current MCP tools do not publish content externally or delete projects and nodes.
MeshWrite Skill for agents
MCP gives an agent the tools; the MeshWrite Skill explains when and how to use them, including the difference between unique node IDs and project references, and the fact that saved answers remain drafts.
Canonical Skill URL:
https://meshwrite.cc/skills/meshwrite/SKILL.mdLatest-version manifest:
https://meshwrite.cc/skills/meshwrite/manifest.jsonTo refresh an agent that already has the Skill, ask it to fetch the manifest and reload the Skill from the canonical URL when the version changes. The fixed URL keeps the address stable; it does not force every AI platform to refresh a cached Skill automatically. Update behavior depends on that platform.
⚡ Evolution Studio & Branching Matrix
The Evolution Studio is where generative AI meets the version tree. When branching nodes, you define the Evolution Matrix in a precise two-step configuration:
- Select AI Role / Persona: Assign custom tasks to your agents, such as the Absolute De-Fluffer (stripping bloated adjectives), First-Principles Analyzer (deconstructing assumptions), or Outliner.
- Pair with Provider & Model: Pair each persona with a specific LLM model configuration. Run lightweight tasks on faster models, and complex reasoning on pro endpoints.
Pipeline Execution: Parallel vs. Serial
When running a multi-agent evolution matrix, you can choose how they interact:
Triggers multiple alternative versions from the exact same parent node simultaneously. Ideal for comparing different hooks, tones, or structures at the same time.
Sequentially chains agent operations. The output of Agent A (e.g. outline generator) is automatically fed as the input to Agent B (e.g. copywriter) to polish and refine.
📥 Web Clipper Setup
The Web Clipper Chrome Extension serves as a lightning-fast collector for your active reading.
Rather than managing bookmarks or linear scrapbooks, highlights and notes are sent instantly into your central Inbox. It bypasses choice fatigue so you can continue your research flow undisturbed.
Download the extension from the Chrome Web Store, log in, and use the shortcut shown by the extension to clip highlighted thoughts.
Processing Inbox Items
When you open the Evolution Studio workspace, the Inbox panel in the main console lists all captured web snippets.
Drag and drop any inbox card directly onto the canvas grids to create a root version node, or connect it as a child draft reference in your current project draft.