Connect your AI agent to human experts via MCP

What Tendem MCP does

Tendem is an MCP server that connects your AI agent to a network of over 10,000 verified domain experts. When your agent encounters a task requiring human judgment, it calls a Tendem tool the same way it would call any other MCP tool. The expert completes the task, and the result comes back as structured JSON.

No SDK to import. No wrapper functions. No code changes to your existing agent. MCP tools are discovered automatically. Your agent decides when to delegate on its own, or you can define explicit escalation rules.

Built by Toloka, the same company trusted by Anthropic, Shopify, and leading AI labs for high-quality human data, Tendem brings the same expert infrastructure into your agent’s runtime workflow. For background on how Model Context Protocol works, see our MCP explainer.

Prerequisites

Tendem account: Sign up at agent.tendem.ai (free to start, no credit card required).

API key: Generate at agent.tendem.ai/tokens. This token authenticates your agent with Tendem’s servers.

MCP client: Any MCP-compatible application: Claude Desktop, Cursor, Claude Code, or a custom agent built with the MCP SDK.

Runtime (for local setup): Python 3.10+ with uv installed, for the uvx command. Not required if using remote MCP transport.

Setup for Claude Desktop

Claude Desktop is the most common MCP client. Setup takes about two minutes.

Step 1. Open Claude Desktop. Go to Settings, navigate to the Developer section, and click "Edit Config." This opens the claude_desktop_config.json file.

Step 2. Add the Tendem server to your configuration:

{

  "mcpServers": {

    "tendem": {

      "command": "uvx",

      "args": ["tendem-mcp"],

      "env": {

        "TENDEM_API_KEY": "your-api-key-here"

      }

    }

  }

}

Step 3. Replace "your-api-key-here" with the API key from agent.tendem.ai/tokens.

Step 4. Restart Claude Desktop. The Tendem tools will appear in your tools menu.

Setup for Claude Code

Claude Code supports adding MCP servers with a single terminal command:

claude mcp add tendem -e TENDEM_API_KEY=your-api-key -- uvx tendem-mcp

Run /mcp inside Claude Code to verify the connection is active and tools are discovered.

Setup for Cursor

Cursor supports MCP servers at both the project level and globally.

Project-level: Create or edit .cursor/mcp.json in your project root:

{

  "mcpServers": {

    "tendem": {

      "command": "uvx",

      "args": ["tendem-mcp"],

      "env": {

        "TENDEM_API_KEY": "your-api-key-here"

      }

    }

  }

}

Global: Edit ~/.cursor/mcp.json with the same configuration.

After saving, go to Cursor Settings, navigate to MCP, and click the refresh button.

Setup for a custom Python agent

If you’re building your own agent using the MCP Python SDK, connect to Tendem programmatically:

import asyncio, os

from agents import Agent, Runner

from agents.mcp import MCPServerStdio

async def main():

    async with MCPServerStdio(

        name="Tendem",

        params={

            "command": "uvx",

            "args": ["tendem-mcp"],

            "env": {"TENDEM_API_KEY": os.environ["TENDEM_API_KEY"]},

        },

    ) as server:

        agent = Agent(

            name="Research Assistant",

            instructions="Use Tendem tools when you need expert verification.",

            mcp_servers=[server],

        )

        result = await Runner.run(agent, "Verify the market size data.")

        print(result.final_output)

asyncio.run(main())

Set your API key as an environment variable before running: export TENDEM_API_KEY=your-api-key

Give your AI agent a human expert

Tendem connects your agent to 20,000+ vetted specialists via MCP, no SDK, no code changes.

Get your API key →

Your first Tendem task: three examples

Example A: fact verification

Your agent researches a topic using web search and database MCP servers, then needs to verify the accuracy of key claims before including them in a deliverable.

The agent calls Tendem’s tool with the claims to verify. Tendem matches the task to a research specialist who checks each claim against primary sources, flags any errors, adds citations, and returns structured JSON with verified data, source URLs, and a quality score. The response includes a confidence rating for each claim and links to primary sources used for verification.

Example B: market research

Your agent identifies a list of companies in a target market segment using automated data sources, but needs deeper analysis that goes beyond what databases provide.

The agent delegates the analysis to Tendem. An analyst with relevant sector experience reviews the data, adds competitive context, identifies patterns the automated sources missed, and delivers a structured research summary. The analyst brings information that automated scrapers cannot: industry knowledge about recent acquisitions, context about misleading growth metrics, and insight into market dynamics.

Example C: data scraping with human verification

Your agent automates web data extraction from dozens of sources, but you need 99%+ accuracy for the resulting dataset. Pure automation typically achieves 85-95% accuracy.

The agent handles the initial extraction via automated MCP tools, then routes the extracted data to Tendem for human verification. Expert reviewers validate accuracy, resolve edge cases, correct misclassifications, and ensure completeness. Common applications include building lead databases, compiling pricing intelligence, and assembling research datasets where data quality directly affects downstream analysis.

How the quality pipeline works

When your agent sends a task to Tendem, it goes through a structured pipeline:

1. Task intake and clarification. Tendem’s AI agent analyzes the request, breaks it into subtasks, and identifies what information is needed.

2. Expert matching and execution. The task is matched to the best-suited specialist based on domain, language, complexity, and track record. AI handles routine steps while the human expert provides oversight and refinement.

3. Automated QA. The deliverable passes through automated quality checks: consistency verification, source validation, format compliance, and completeness scoring.

4. Human QA review. A separate QA specialist reviews the output against the original requirements.

5. Structured delivery. The final result is returned as structured JSON containing the verified deliverable, source citations, quality score, and metadata. The pipeline is fully transparent: you can see what AI handled, what the expert contributed, and what QA flagged.

Tips for effective integration

Define escalation rules. Don’t rely solely on the agent’s judgment. Set explicit rules: always verify financial data, always get expert review on compliance content, always validate before publishing.

Use structured requests. Include context about what the data will be used for, what accuracy standards apply, and what format you need.

Leverage non-blocking workflows. Design your agent to continue processing other tasks while waiting for Tendem responses.

Test with MCP evaluations. Run trajectory-based evaluations to verify your agent correctly identifies when to escalate.

Give your AI agent a human expert

Tendem connects your agent to 20,000+ vetted specialists via MCP, no SDK, no code changes.

Get your API key →

Frequently asked questions

How long does a Tendem expert response take?

What domains do Tendem experts cover?

Is my data secure when using Tendem MCP?

Can I use Tendem with non-Anthropic models?

How much does Tendem cost?