MCP Server

The IntentText MCP server gives LLMs direct access to parsing, rendering, querying, trust operations, and template merging through the Model Context Protocol.

Installation

npm install -g @dotit/mcp

Configuration

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "intenttext": {
      "command": "intenttext-mcp"
    }
  }
}

Or run it without a global install via npx:

{
  "mcpServers": {
    "intenttext": {
      "command": "npx",
      "args": ["-y", "@dotit/mcp"]
    }
  }
}

VS Code (Copilot)

Add to your .vscode/mcp.json:

{
  "servers": {
    "intenttext": {
      "command": "intenttext-mcp"
    }
  }
}

HTTP mode

For remote or shared setups, the package ships a second binary that serves the same tools over Streamable HTTP:

PORT=3847 intenttext-mcp-http

PORT defaults to 3000.

Available tools

All tools operate on IntentText source strings — pass the .it text in, get JSON or rendered output back.

Parse & serialize

Tool	Parameters	Description
parse_intent_text	source, safe	Parse .it source into a structured JSON document. safe (default true) collects warnings instead of throwing
document_to_source	document	Convert a document JSON object back to .it source text

Render

Tool	Parameters	Description
render_html	source, theme?	Render to styled HTML. Optional built-in theme: corporate, minimal, warm, technical, print, legal, editorial, dark
render_print	source, theme?	Print-optimised HTML with @media print CSS — applies font: and page: blocks; suitable for PDF generation. Same optional theme

Templates

Tool	Parameters	Description
merge_template	template, data, render?, missing?	Merge {{variable}} placeholders with a data object. render: none (default, returns .it source), html, or print. missing: keep leaves unresolved placeholders visible (default), blank removes them so optional fields never print

Validate & query

Tool	Parameters	Description
validate_document	source	Semantic validation beyond syntax: broken step references, missing required properties, unresolved variables, workflow logic errors
query_document	source, type?, content?, section?, limit?	Filter blocks by type (comma-separated for multiple: step,gate), content substring, or section
diff_documents	before, after	Semantic diff between two versions — which blocks were added, removed, or modified

Workflow

Tool	Parameters	Description
extract_workflow	source	Extract the execution graph: steps in topological order, dependencies, parallel batches, gate positions

Trust

Tool	Parameters	Description
seal_document	source, signer, role?	Seal with a cryptographic signature — appends sign: and freeze: blocks, returns sealed source + hash
verify_document	source	Verify integrity of a sealed document — reports tampering
get_document_history	source, block_id?, section?	Revision history of a tracked document: what changed, who, when

Tool examples

Parse and render

User: "Render this invoice as HTML with the corporate theme"

Claude calls: render_html({
  source: "title: Invoice #2847\n...",
  theme: "corporate"
})

Merge a template

User: "Generate the invoice for Acme from the standard template"

Claude calls: merge_template({
  template: "title: Invoice {{number}}\n...",
  data: { "number": "INV-2847", "client": "Acme Corp" },
  render: "print",
  missing: "blank"
})

Query a document

User: "What tasks are in this project plan?"

Claude calls: query_document({
  source: "title: Project Plan\n...",
  type: "task",
  limit: 20
})

Seal a contract

User: "Seal this NDA as Maria Santos, COO"

Claude calls: seal_document({
  source: "title: NDA\n...",
  signer: "Maria Santos",
  role: "COO"
})

The result includes the sealed source — store it exactly as returned (the hash covers the exact bytes).

Validate a generated workflow

User: "Check this deployment workflow before we run it"

Claude calls: validate_document({
  source: "step: Build\nstep: Deploy | after: Build\n..."
})

Security

The MCP server is stateless: every tool takes document source as input and returns results — it does not read or write your filesystem. Trust operations (seal, verify, history) use the same SHA-256 integrity system as the rest of the toolchain.

Source

Repository: intenttext-mcp · npm: @dotit/mcp (1.0.0 — formerly @intenttext/mcp, now deprecated with a pointer)