What is MCP and why does the Hermes API support it?

MCP (Model Context Protocol) is the open standard for connecting LLM clients like Claude Desktop, Cursor, and Windsurf to tools and data. Hermes ships an MCP server so the same agent you call over REST is available as a native tool in any MCP client.

Is the REST API stable enough for production?

Yes. All /v1/ endpoints are versioned with a 6-month deprecation policy, idempotency keys, retries, and signed webhooks.

Can I host the agent runtime myself?

Yes. Hermes is built on OpenClaw, which is open source. Self-host the runtime and point it at Donely's control plane, or run fully air-gapped on Enterprise.

Hermes API · MCP · CLI · SDKs

Hermes Agent as an API

Run, orchestrate, and observe Hermes agents from any client. REST endpoints, an MCP server for Claude/Cursor/Windsurf, a first-class CLI, typed SDKs, webhooks, and streaming tokens — all backed by the same managed runtime.

Get an API key Quickstart

bash

curl https://api.donely.ai/v1/hermes/runs \
  -H "Authorization: Bearer $DONELY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"agent":"my-agent","input":"Summarize today\'s sales calls"}'

Hermes is built for the new agent stack — where the same agent has to answer a REST call, plug into Claude as an MCP tool, run inside a CI pipeline via the CLI, and stream tokens into your product UI. One agent, every surface.

What you can do

Every surface agents need — one platform

REST, MCP, CLI, SDKs, webhooks, and streaming. Wire Hermes into anything from a Cursor session to a production backend.

REST API

Create runs, stream events, manage agents, knowledge, and tools over standard HTTPS with bearer auth.

MCP Server

Drop the Hermes MCP URL into Claude Desktop, Cursor, Windsurf, or any MCP-aware client. Tools, prompts, and resources exposed natively.

CLI

Trigger agents, tail logs, deploy skills, and pipe agent output into shell scripts and CI jobs.

Typed SDKs

TypeScript and Python SDKs with full type coverage, streaming helpers, retries, and idempotency keys.

Webhooks

Subscribe to run.completed, run.failed, tool.called, and message.sent events with HMAC-signed payloads.

Streaming

Server-sent events for live token streaming, tool calls, and intermediate reasoning steps.

Guardrails

Per-key rate limits, spend caps, allow-lists for tools, and audit logs on every call.

OAuth & Keys

Personal API keys for prototyping, scoped service keys for backends, OAuth for end-user agents.

Long-running runs

Background runs with resumable state, idempotency, and replay — built for jobs that take minutes, not milliseconds.

Quickstart

Pick your surface

The same Hermes agent runtime, addressable from REST, MCP, the CLI, or a typed SDK.

bash

# Create a run
curl https://api.donely.ai/v1/hermes/runs \
  -H "Authorization: Bearer $DONELY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "sales-research",
    "input": "Find recent funding news for Acme Corp",
    "stream": true
  }'

# Tail the event stream
curl -N https://api.donely.ai/v1/hermes/runs/run_123/events \
  -H "Authorization: Bearer $DONELY_API_KEY"

REST base

https://api.donely.ai/v1/hermes

MCP endpoint

https://mcp.donely.ai/hermes/sse

CLI

npm i -g @donely/hermes-cli

What people build

From IDE copilots to production backends

MCP tool in Claude or Cursor

Expose a Hermes agent (with your CRM, docs, and tools) to your IDE or chat client in 30 seconds.

Backend automation

Call Hermes from a serverless function to triage tickets, draft replies, or enrich leads on every webhook.

CLI-driven workflows

Pipe agent output through shell scripts and CI — replace brittle prompt-glue scripts with a real agent.

Product-embedded agents

Stream tokens straight into your app UI via SSE, with structured tool events and citations.

Scheduled jobs

Long-running, resumable runs that fit into cron — daily research, weekly summaries, nightly reconciliations.

Per-user OAuth agents

Issue scoped keys per end-user so your SaaS can give every customer their own Hermes instance.

FAQ

Questions, answered honestly

MCP (Model Context Protocol) is the emerging open standard for connecting LLM clients like Claude Desktop, Cursor, and Windsurf to external tools and data. Hermes ships an MCP server out of the box, so the same agent you call over REST is instantly available as a native tool inside any MCP-aware client — no plugin code to write.

Yes. All endpoints under /v1/ are versioned and follow a deprecation policy: 6 months notice before any breaking change, with parallel versions while you migrate. Idempotency keys, retries, and webhook signatures are first-class.

The CLI is for humans, shell scripts, and CI — quick runs, tailing logs, deploying skills. The SDKs are for app code with type safety, streaming helpers, and connection pooling. They share the same underlying API.

Yes. Hermes is built on OpenClaw, which is open source. You can self-host the runtime and still get the managed dashboard, billing, and observability by pointing it at Donely's control plane — or run fully air-gapped on Enterprise.

Runs are durable: state is persisted, runs survive worker restarts, and you can subscribe to the same run's event stream from multiple clients. Webhooks fire on completion so you don't need to poll.

Per-key rate limits and spend caps are configurable. Pricing is based on agent compute time + LLM tokens — bring your own keys to pay providers directly, or use included credits on paid plans.

Your AI workforce is one click away

Get started in minutes. No setup, no infrastructure headaches.

Deploy your first AI employee

Free $5 credit · No credit card required · 60s setup