Claude Desktop was the first AI assistant to offer native MCP integration. The current model family includes:
This guide covers setting up Tallyfy’s MCP server with Claude Desktop, what works well, and where to expect limitations.
Key points:
Important: Claude Desktop vs Claude Computer Use
This article covers Claude Desktop’s text-chat MCP integration - natural language access to Tallyfy data via API. This differs from Claude Computer Use, which controls computer interfaces visually (mouse, keyboard, screenshots).
Key differences:
Use MCP for data queries, analysis, and API automation. Use Computer Use for visual UI tasks that need clicking and seeing interface elements.
Claude Desktop’s MCP features include:
Install Claude Desktop
Download from claude.ai/download ↗ for macOS or Windows.
Find the configuration file
~/Library/Application Support/Claude/claude_desktop_config.json%APPDATA%\Claude\claude_desktop_config.jsonClaude creates this file when you first edit the configuration.
Create the Tallyfy MCP server
mkdir ~/tallyfy-mcp-servercd ~/tallyfy-mcp-servernpm init -ynpm install @modelcontextprotocol/sdk axiosWrite the server script
Create tallyfy-server.js:
const { Server } = require('@modelcontextprotocol/sdk/server/index.js');const { StdioServerTransport } = require('@modelcontextprotocol/sdk/server/stdio.js');const axios = require('axios');
const TALLYFY_API_KEY = process.env.TALLYFY_API_KEY;const TALLYFY_API_URL = 'https://mcp.tallyfy.com';
const server = new Server({ name: 'tallyfy-mcp', version: '1.0.0',});
server.setRequestHandler('tools/call', async (request) => { if (request.params.name === 'search_tasks') { const { query, status } = request.params.arguments;
const response = await axios.get(`${TALLYFY_API_URL}/tasks`, { headers: { 'Authorization': `Bearer ${TALLYFY_API_KEY}` }, params: { q: query, status } });
return { content: [{ type: 'text', text: JSON.stringify(response.data, null, 2) }] }; }});
server.setRequestHandler('tools/list', async () => { return { tools: [{ name: 'search_tasks', description: 'Search for tasks in Tallyfy', inputSchema: { type: 'object', properties: { query: { type: 'string', description: 'Search query' }, status: { type: 'string', enum: ['open', 'completed', 'all'] } } } }] };});
const transport = new StdioServerTransport();server.connect(transport);Configure Claude Desktop
Option A: Desktop Extensions (recommended)
Option B: Manual configuration
Edit claude_desktop_config.json:
{ "mcpServers": { "tallyfy": { "command": "node", "args": ["/Users/username/tallyfy-mcp-server/tallyfy-server.js"], "env": { "TALLYFY_API_KEY": "your-tallyfy-api-key-here" } } }}Replace /Users/username/ with your home directory and add your Tallyfy API key.
Restart Claude Desktop
Fully quit (Cmd+Q on macOS) and relaunch.
Verify the connection
Look for the hammer icon at the bottom of the chat. Click it to see your Tallyfy tools.
Test with:
"Search for all open tasks in Tallyfy"Find all tasks assigned to me due this week and create a summary report.Claude will query Tallyfy via the search_tasks tool, filter by user and date range, format a structured report, and optionally save it locally via filesystem MCP.
Check for stalled processes in our "Customer Onboarding" template and suggest next actions.Claude queries active processes, identifies tasks that haven’t progressed, analyzes blockers, and provides recommendations.
Reassign all of John's open tasks to Sarah - he's on vacation.Claude finds John’s tasks, filters for open status, reassigns them, and summarizes the changes.
Claude Desktop’s Projects feature lets you create a project per workflow or client. Upload process docs once and reference them across sessions. Task history persists between conversations - ideal for managing processes that span weeks or months.
Projects also support memory - Claude extracts and saves key facts to maintain continuity. Extended thinking lets it work through multi-step problems, alternating between reasoning and tool calls.
Unlike web-based AI, Claude Desktop can access your local files:
{ "mcpServers": { "filesystem": { "command": "npx", "args": [ "-y", "@modelcontextprotocol/server-filesystem", "/Users/username/Documents/Tallyfy" ] } }}This lets you read process docs from disk, export Tallyfy data to CSV, and sync local templates.
Connecting multiple MCP servers is where Claude Desktop gets powerful. Example multi-server workflow:
~/Library/Logs/Claude/ (macOS) or %APPDATA%\Claude\logs\ (Windows)Claude Desktop is text-only. You can’t see Tallyfy’s visual process tracker, clickable flowcharts, or progress bars. You’ll get “75% complete” as plain text instead.
Workaround: Have Claude pull the data, then view it in Tallyfy’s web UI.
MCP is request-response, not live streaming. There are no push notifications when tasks change - you must ask Claude to check.
Workaround: Schedule periodic checks or set up webhooks for urgent items.
Rich forms become flat text. Multi-select dropdowns with 50 options get listed out. File uploads need a separate step. Date pickers become manual text entry.
One API key means one identity. Everything runs as the key owner, you can’t switch accounts, and the audit trail shows a single person for all actions.
Show me all tasks due today, any overdue items, and processes needing my attention. Create a prioritized action list.Analyze the last 50 completed 'Customer Onboarding' processes. Find the steps with longest completion times and suggest improvements.Generate an SOP document for our 'Invoice Processing' template with all steps, responsible parties, and average completion times.Review team member skills and current workload, then suggest optimal task assignments for next week's projects.Create a compliance report showing all 'Data Access Request' processes from last quarter, including completion times and SLA violations.Pair Claude’s text-based MCP with Claude Computer Use for broader automation:
MCP handles: API data queries, bulk operations, report generation, data-driven decisions
Computer Use handles: Visual interfaces, third-party form filling, UI navigation, apps without APIs
Step 1 - MCP (data gathering):
Analyze all completed audit processes this month, identify non-compliance issues, and export a summary with statistics to CSV.Step 2 - Computer Use (visual automation):
Open the government compliance portal, navigate to the monthly report section, and fill in the form using the CSV data. Screenshot each step for audit trail.Configuration security
"sensitive": true in manifest.json for auto-encryptionchmod 600 claude_desktop_config.jsonMCP server isolation
https://go.tallyfy.com/mcp/oauth/authorizehttps://go.tallyfy.com/mcp/oauth/tokenhttps://go.tallyfy.com/mcp/oauth/register/.well-known/oauth-authorization-server (RFC 8414)mcp.{resource}.{action} - for example: mcp.tasks.read, mcp.tasks.write, mcp.processes.read, mcp.templates.write, mcp.users.read, mcp.forms.read, mcp.automation.writeData handling
~/Library/Application Support/Claude/Network security
Claude Desktop supports multiple simultaneous MCP connections:
{ "mcpServers": { "tallyfy": { "command": "node", "args": ["~/tallyfy-mcp-server/server.js"], "env": { "TALLYFY_API_KEY": "key-1" } }, "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "~/Documents"] }, "github": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-github"], "env": { "GITHUB_TOKEN": "ghp_xxx" } } }}Use TypeScript for type safety
import { Server } from '@modelcontextprotocol/sdk/server/index.js';import { TallyfyClient } from './tallyfy-client';Handle errors properly
try { const result = await tallyfyApi.call(); return { content: [{ type: 'text', text: result }] };} catch (error) { return { error: { code: 'TALLYFY_API_ERROR', message: error.message } };}Add request logging
console.error(`[Tallyfy MCP] ${new Date().toISOString()} - ${request.method}`);/ or double backslashes \\ in JSON configsClaude Desktop:
ChatGPT/OpenAI:
Choose Claude for the most mature MCP experience, Desktop Extensions, Projects for persistent context, and Computer Use hybrid workflows. Consider ChatGPT for Microsoft integration or enterprise admin-controlled connectors.
inputSchema definitionshttps://mcp.tallyfy.com using OAuth 2.1 with PKCE so you can manage tasks and processes and templates and automations through natural language in both Claude.ai web and Claude Desktop without writing any API calls. Mcp Server > Using Tallyfy MCP server with ChatGPT
Mcp Server > Using Tallyfy MCP server with Microsoft Copilot Studio