Skip to main content

Cursor

Edit ~/.cursor/mcp.json (or open Cursor settings → MCP → edit config):

Hosted (recommended)

{
  "mcpServers": {
    "talonic": {
      "url": "https://mcp.talonic.com/mcp",
      "headers": {
        "Authorization": "Bearer tlnc_your_key_here"
      }
    }
  }
}

Local (npx)

{
  "mcpServers": {
    "talonic": {
      "command": "npx",
      "args": ["-y", "@talonic/mcp@latest"],
      "env": {
        "TALONIC_API_KEY": "tlnc_your_key_here"
      }
    }
  }
}

After saving the config, restart Cursor or reload the window. The Talonic tools appear in the MCP tool list and can be invoked from Cursor's AI chat panel. Cursor's agent mode is particularly effective with Talonic because it can read local files from your project and pass them directly to talonic_extract via file_path.

A common Cursor workflow is extracting data from documents within a codebase — for example, parsing specification PDFs, extracting configuration from scanned documents, or converting legacy documentation to structured data. The file_path input works directly with paths relative to your Cursor workspace.

Cursor caches MCP tool descriptions between sessions. If you upgrade the MCP server and notice outdated tool descriptions, restart Cursor to force a refresh of the tool metadata.

Example: extracting from a project file

Agent extracts data from a spec PDF in the workspace
// User: "Extract the requirements from docs/product-spec.pdf"
// Cursor agent calls talonic_extract with:
{
  "file_path": "./docs/product-spec.pdf",
  "schema": {
    "type": "object",
    "properties": {
      "title": { "type": "string" },
      "requirements": {
        "type": "array",
        "items": {
          "type": "object",
          "properties": {
            "id": { "type": "string" },
            "description": { "type": "string" },
            "priority": { "type": "string" }
          }
        }
      },
      "deadline": { "type": "string", "format": "date" }
    }
  }
}

Cursor's composer mode is particularly powerful with Talonic because the agent can extract data from a document and immediately use it to generate code. For example, you can ask Cursor to extract fields from an API specification PDF and then generate TypeScript interfaces matching the extracted schema. The agent chains talonic_extract with code generation in a single turn, eliminating manual data entry.

When using the hosted MCP option in Cursor, the connection is established over HTTPS with no local process. This means Cursor starts faster and there is no Node.js dependency. For teams sharing a Cursor workspace configuration, the hosted option is simpler to standardise because every developer only needs an API key — no local Node.js version management needed.

Cursor supports both the hosted and local MCP options. The hosted option is simpler since Cursor can connect directly via HTTPS without spawning a local process.

Frequently asked questions

How do I add Talonic MCP to Cursor?+
Edit ~/.cursor/mcp.json and add the Talonic MCP server config with your API key.
Can Cursor read local files for extraction?+
Yes. Cursor's agent mode can read files from your project and pass them to talonic_extract via the file_path parameter, making it easy to extract data from documents within your codebase.
Why are Talonic tool descriptions outdated in Cursor?+
Cursor caches MCP tool descriptions between sessions. Restart Cursor to force a refresh after upgrading the MCP server.
Can Cursor chain extraction with code generation?+
Yes. In composer mode, Cursor can extract data from a document and immediately use the results to generate code — for example, extracting fields from an API spec PDF and generating TypeScript interfaces. The agent handles both steps in a single turn.

Related

Claude Desktop
Cline