Use Pathbound in any agent that speaks MCP.

The pages here cover the products we have tested, but Pathbound is one remote MCP server, so any MCP-capable agent can read the same customer profiles. Point a custom or in-house agent, or a runtime we have not documented yet, at the endpoint and it gets the resolved contact, company, timeline, and conversation history.

Why connect it

One endpoint, any MCP client.

If your agent can call a remote MCP server, it can read Pathbound. There is no Pathbound-specific SDK to install.

Connect

Connect Pathbound to a custom agent.

Two paths: a bearer API key for a headless or in-house runner, or interactive OAuth for a client that can open a browser.

  1. Create a scoped Pathbound key

    In Pathbound, open REST API keys and create a key with contacts:read, companies:read, and events:read. Add write scopes later, only when a workflow needs them.

  2. Point your MCP client at the endpoint

    Register Pathbound as a remote Streamable HTTP MCP server and send the key in the Authorization header. Most MCP client libraries take a URL and a set of headers.

    MCP client config
    {
      "mcpServers": {
        "pathbound": {
          "type": "http",
          "url": "https://mcp.pathbound.ai/mcp",
          "headers": { "Authorization": "Bearer ${PATHBOUND_API_KEY}" }
        }
      }
    }
  3. Or complete OAuth for an interactive client

    If the client supports MCP OAuth, add the same URL https://mcp.pathbound.ai/mcp and let it authorize in the browser. No API key is stored in that case.

  4. List the tools, then read one known customer

    Confirm the client discovered the Pathbound tools, then ask it to read a contact you can verify (for example get_contact then get_contact_timeline) before wiring it into a workflow.

New to MCP? The MCP docs cover the endpoint, tools, and authentication. Popular agent SDKs (the OpenAI Agents SDK, the Vercel AI SDK, Google ADK, and others) all accept a remote MCP server.

Access

Keep the first connection read-only.

A custom agent can run unattended. Decide what it can read before you let it write.

In Pathbound

Create the smallest key scope that serves the job, and rotate or revoke it from the dashboard. A key without actions:write receives no write or action tools.

In your agent

Store the key in the runtime environment, never in source control. Start read-only, add one write path at a time, and keep an approval step around anything that changes a record.

Custom setups and Pathbound.

Which agents can connect?

Any client or SDK that can call a remote MCP server over Streamable HTTP, including custom in-house agents and runtimes we have not written a page for yet. If your client supports MCP, it supports Pathbound.

OAuth or an API key?

Use interactive OAuth when the client can open a browser and each person should connect their own Pathbound identity. Use a scoped sk_ API key as a bearer token for a headless or shared runner.

Do I need a Pathbound SDK?

No. Pathbound is a standard MCP server, so your existing MCP client library is enough. Send a URL and an Authorization header; Pathbound returns the tools.

Want a guide for a specific runtime?

If you would like a step-by-step page for a product we have not covered, tell us. The documented runtimes are on the use-with hub; the rest connect the same way as above.

Can a custom agent write back?

Only when the key includes the required write scopes and the connected source allows the tool. Keep the first setup read-only. Governed write-back is rolling out; versioning and rollback are not live yet.

Related
Try it on your own data

Within five minutes, your agent is querying real customer context.

Sign up, connect one CRM or drop the events snippet, and point your MCP client at mcp.pathbound.ai/mcp.