Connect AI Agents to PacedLoop via Model Context Protocol (MCP)

A clear setup flow, client-specific config blocks, troubleshooting guidance, and a tighter endpoint decision path for connecting agents to PacedLoop over HTTP.

New to PacedLoop? See how the platform works.

Get tokens Jump to configs

What agents can doNatural language to tools

“Start a new run for the onboarding workflow.”
“Submit the user's answer for step 2 and tell me the next step.”
“Show me recent runs for this workflow.”
“Update the tone of step 3 in the hosted workflow.”

High level steps

Generate the right token

Open Workspace -> API Token and create either a workflow run token for /api/mcp-runs or an MCP token for /api/mcp. Store it immediately and keep it server-side or in a secure local config.

Add the server to your AI tool

Use an HTTP MCP entry with an Authorization header. Most tools only need a server URL plus Bearer token, while Claude Desktop and Claude Code use an HTTP bridge through npx mcp-remote.

Restart and verify tools appear

After saving the config, fully restart the app or CLI session. If tools do not appear, validate the config file syntax, confirm the exact endpoint, and check for a clean Bearer header with a space before the token.

MCP Server selection by use case

If the client is mainly consuming an existing workflow, use the workflow run MCP. If the client belongs to the workflow owner and should also create, edit, or analyze workflows, use the hosted MCP.

Narrow surface

Workflow run MCP

A run-focused MCP surface for people and agents that should execute an existing workflow without getting workflow management or account-wide visibility.

Endpoint: https://www.pacedloop.com/api/mcp-runs
Token: Workflow run token

Best for

A workflow creator shares a workflow run token or API key and the end user connects their own MCP client
A custom GPT, chatbot, or coding agent that should only run a predefined workflow
External users of a workflow who should not create, edit, or inspect the broader account

Main tools

run_startrun_advancerun_repairrun_statusrun_get_step

This uses the same workflow run token as your customGPT's.

Broader surface

Hosted MCP

The full hosted surface for workflow management, run execution, run history, and analytics.

Endpoint: https://www.pacedloop.com/api/mcp
Token: MCP token

Best for

Workflow owners who want one MCP server for both building workflows and running them
Agents that create, update, or delete workflows and then execute them
Internal automations that also need run lists, full run records, or analytics data

Main tools

workflow_listworkflow_getworkflow_createworkflow_updateworkflow_step_updaterun_listrun_getanalytics_overview

The workflow run token is rejected here. Use a dedicated MCP token.

Tool Selection by Outcome

Run execution

The runs-only endpoint is built for agents that move a workflow forward one step at a time.

run_startrun_advancerun_statusrun_get_steprun_repair

Workflow management

The hosted endpoint adds workflow definition access for creation, editing, and deletion.

workflow_listworkflow_getworkflow_step_getworkflow_createworkflow_updateworkflow_step_updateworkflow_delete

Operational visibility

Hosted MCP also exposes run history and analytics for monitoring usage and diagnosing issues.

run_listrun_getanalytics_overviewanalytics_recent_runsanalytics_steps

Tool setup

Cursor

~/.cursor/mcp.json or .cursor/mcp.json

{
  "mcpServers": {
    "pacedloop": {
      "url": "https://www.pacedloop.com/api/mcp-runs",
      "headers": {
        "Authorization": "Bearer <<YOUR_WORKFLOW_RUN_TOKEN>>"
      }
    }
  }
}

VS Code / GitHub Copilot

.vscode/mcp.json or user-level mcp.json

{
  "servers": {
    "pacedloop": {
      "type": "http",
      "url": "https://www.pacedloop.com/api/mcp-runs",
      "headers": {
        "Authorization": "Bearer <<YOUR_WORKFLOW_RUN_TOKEN>>"
      }
    }
  }
}

Claude Desktop / Claude Code

Claude config file or claude mcp add

{
  "mcpServers": {
    "pacedloop": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://www.pacedloop.com/api/mcp-runs",
        "--header",
        "Authorization: Bearer ${PACEDLOOP_WORKFLOW_RUN_TOKEN}"
      ],
      "env": {
        "PACEDLOOP_WORKFLOW_RUN_TOKEN": "paste-your-workflow-run-token-here"
      }
    }
  }
}

OpenAI Codex

~/.codex/config.toml

[mcp_servers.pacedloop]
url = "https://www.pacedloop.com/api/mcp-runs"
headers = { Authorization = "Bearer <<YOUR_WORKFLOW_RUN_TOKEN>>" }

Runs-only JSON starter

{
  "mcpServers": {
    "pacedloop": {
      "url": "https://www.pacedloop.com/api/mcp-runs",
      "headers": {
        "Authorization": "Bearer <<YOUR_WORKFLOW_RUN_TOKEN>>"
      }
    }
  }
}

Hosted MCP swap

{
  "mcpServers": {
    "pacedloop": {
      "url": "https://www.pacedloop.com/api/mcp",
      "headers": {
        "Authorization": "Bearer <<YOUR_MCP_TOKEN>>"
      }
    }
  }
}

The failures are usually config shape, token mismatch, or missing local bridge tooling.

Tools are not showing up

Restart the client fully, then validate the config file syntax. If the client reads JSON, remove trailing commas and confirm the server key nesting is correct.

Authentication errors

Check that the header is exactly Bearer <token>, with a space after Bearer. Also confirm that you are using the workflow run token for /api/mcp-runs and the MCP token for /api/mcp.

Claude bridge errors

If Claude Desktop or Claude Code cannot start the server, verify that Node.js is installed and that npx is available in your terminal environment.

Generate the correct token before wiring the server into your client.

Workflow run token for /api/mcp-runs. Dedicated MCP token for /api/mcp. Both are available from the workspace token page.

Open API Token page Workspace docs version