Design
zindex.ai
An MCP server for creating, retrieving, validating, and version-editing saved DSP or Mermaid diagrams.
ENDPOINT 1
https://api.zindex.ai/mcp
MCP server metadata
- Name
- zindex
- Version
- 6.0.0+build.1824
Immutable agent diagrams. Zindex keeps every saved version exactly, so a diagram made in one conversation can be found, changed and rendered the same way later, from any MCP client. Loop: diagram_find or diagram_create, then diagram_get, diagram_edit, diagram_render. Edits need baseRevision (the revision you last read) and a one-line message. - diagram_get returns a compact outline. Pass knownRevision to get only what changed since then. - Mermaid works both ways for erDiagram and flowchart: diagram_create takes it, diagram_get view mermaid returns it, and diagram_edit takes the edited text (architecture and ER diagrams) and applies only the differences. Element IDs are the Mermaid IDs. - Retries are safe: use key on diagram_create and requestId on diagram_edit. - On REVISION_CONFLICT the error lists what changed; re-plan against the new revision. - Keep a brief of up to 1,000 characters on each diagram: what it shows, conventions to keep, decisions made. The next agent reads it first. - Element IDs are yours and stable; reuse them in ops. Skill: https://zindex.ai/skill/SKILL.md. Schemas: https://zindex.ai/schemas/scene.schema.json and https://zindex.ai/schemas/ops.schema.json. - Auto-layout is the default: leave layout off nodes and the engine places them and routes edges (text and image elements need a layout). Nodes keep their positions after the first save unless a node that grows needs the room or an edit changes their frame; new nodes go next to their neighbours, and sequence diagrams follow message order. An autoLayout op lays out again; without elements it also drops most frames' own layout. - diagram_render returns links, not images; preview adds an image. - In a code repository, save the diagram's id and revision in .zindex.json so the next session can pick it up. - When the docs and the engine disagree, or something you needed is missing, tell Zindex with diagram_share action feedback; no diagram is needed, and the user confirms it.
Known tools 9
diagram_findLists this account's diagrams, most recently changed first, with id, key, title and latest revision.
Inferred read-onlydiagram_createSaves a new diagram from a DSP scene or Mermaid text and returns its id, revision 1, content hash and chain hash.
Inferred read-onlydiagram_renderReturns SVG and PNG links to a version of a saved diagram, not the image, with its hashes and engine version.
Inferred read-onlydiagram_historyLists a diagram's versions, newest first, with time, client, message, chain hash and counts of elements added, changed and removed.
Potential side effectsdiagram_sharepublish makes a diagram readable by anyone at its link; unpublish makes it private.
Potential side effectsCONNECT WITH APPROVAL
Client installation
Review this server and its permissions before adding it. Secret placeholders must be set locally.
Codex
~/.codex/config.toml
[mcp_servers.zindex]
url = "https://api.zindex.ai/mcp"
enabled = true
Claude Code
.mcp.json
{
"mcpServers": {
"zindex": {
"type": "http",
"url": "https://api.zindex.ai/mcp"
}
}
}
Claude Desktop
Settings → Connectors → Add custom connector
Name: zindex
Remote MCP URL: https://api.zindex.ai/mcp
Add this remote URL as a custom connector in Claude Desktop. Availability depends on the user plan and workspace policy.
Cursor
.cursor/mcp.json
{
"mcpServers": {
"zindex": {
"url": "https://api.zindex.ai/mcp"
}
}
}
Visual Studio Code
.vscode/mcp.json
Add to Visual Studio Code{
"servers": {
"zindex": {
"type": "http",
"url": "https://api.zindex.ai/mcp"
}
}
}
Generic MCP
Client-specific MCP configuration
{
"name": "zindex",
"transport": "streamable-http",
"url": "https://api.zindex.ai/mcp"
}
MCP Inspector
Run the official MCP Inspector locally and enter the indexed Streamable HTTP endpoint.
TRUST AND VERIFICATION EVIDENCE
Loading Trust v2 evidence…
Checking the associated registrable domain. The BuiltWith key remains server-side.
Evidence is source-attributed and does not guarantee that a third-party server is safe. Risk labels are conservative metadata heuristics.