Agent harnesses

Connect Letta

Letta (ex-MemGPT) · Apache-2.0

SupportedAPI key

Stateful-agent platform with persistent memory in the MemGPT lineage — remote MCPs register per server through the SDK or the ADE.

1 · Choose the MCPs to connect

2 · How should it authenticate?

The config uses a mk_… placeholder — mint your key from MCP Access & keys and paste it in locally. Keys are revocable instantly.

3 · Python SDK

client.mcp_servers.create(
    server_name="deepinsights-mospi",
    config={
        "mcp_server_type": "streamable_http",
        "server_url": "https://marketplace-gateway-pef65a33ta-el.a.run.app/mospi/mcp",
        "auth_token": "Bearer mk_…",
    },
)
  1. 1.Replace mk_… with your personal API key.

4 · Create it as an app

Save this harness + MCP setup to your account. You get a workbench with a live test chat over the combined tool surface — the exact gateway endpoints, auth and tools Letta will use, metered to your own quota.

5 · Or test each MCP on its own

Each playground's Chat and Tools pages exercise exactly the tools your harness will see — same servers, same gateway. Verify a query there first, then run it from Letta.

Agent harnesses