Agent
Configure the agent once. Share it anywhere it needs to work.
An Emcy agent is the runtime boundary for your AI feature. Configure the model, tools, identity, budgets, and embed settings once, then run that same agent internally, share it with customers, or drop it directly into your product.
In-product workflow
The agent lives inside the page your user already trusts
“Emcy turned our OpenAPI spec into a production MCP server in under ten minutes. The OAuth layer just worked.”
Engineering Lead
Series B SaaS
“We went from hosting our own MCP server to Emcy and cut our agent infrastructure cost by 60%.”
CTO
Developer Tools Startup
“The embedded agent SDK let us ship AI features without building a separate integration layer.”
Product Engineer
Enterprise Platform
One configured agent
Share the same agent with your team, your customers, or your product UI.
The point is not another chat demo. The point is one governed agent definition that can power internal workflows, customer-facing help, or a deeply embedded product experience without rebuilding the behavior from scratch for each surface.
- Reuse one agent across internal, partner, and in-product surfaces.
- Keep prompts, tools, and behavior aligned instead of cloning fragile variants.
- Choose where the agent shows up without redefining what the agent is.
In-product workflow
The agent lives inside the page your user already trusts
Configuration
An agent is more than a prompt. It is the policy for how your AI should behave.
Agents are where you decide which tools are connected, which origins are allowed, how the model is framed, and how aggressively the system can spend on behalf of a user. This is the contract your product ships.
- Attach the MCP servers and client tools this agent is allowed to use.
- Control origins, identity behavior, and runtime defaults in one place.
- Give product, engineering, and support one shared definition of the agent.
Product UI
Fill forms, navigate views, and leave visible completion state
Customer profile
Values filled from authenticated API responses
navigate(/settings/profile)
client-side
fill(4 fields)
from API response
showToast("Profile updated")
visible completion
Inside the product
Embed the agent when the best experience is the one that never leaves your app.
When the agent belongs in the product, it should feel native. The agent shows up signed in as the right user, on the right page, with server tools and UI actions available in the same run.
- The agent finishes the task inside the interface, not in a paragraph about it.
- Mix hosted tools with client-side actions so the user sees the result immediately.
- Keep the experience in context instead of opening another tab or dashboard.
In-product workflow
The agent lives inside the page your user already trusts
Per-user spend
Budget the agent like a real production surface, not an experiment.
Once an agent is helping real users, you need to know who triggered it, which tools it touched, and how much it cost. Agent-level controls are how you make AI safe enough to keep on.
- Track usage and spend per user, per agent, and per attached tool surface.
- Set caps so one enthusiastic customer does not become your month-end surprise.
- Use the same observability whether the agent runs internally or in your app.
Total spend
$27.06
of $50.00 budget
Sessions
130
across 4 active users
Budget pressure
1 user
above 90% of quota
| User | Sessions | Spend | Quota |
|---|---|---|---|
SK Sarah Kim Healthy usage | 34 | $7.42 | 74% |
MR Marco Reyes Healthy usage | 15 | $3.18 | 32% |
PP Priya Patel Near limit | 47 | $9.81 | 98% |
Tooling discipline
Keep the prompt small enough that the model can still think.
Every turn does not need every tool. We narrow the prompt to the handful the agent actually needs before the model sees a single token, so the same agent can stay fast and reliable as your catalog grows.
- Smaller prompts mean lower cost and lower latency.
- The agent is less likely to reach for the wrong tool.
- You can grow the tool catalog without turning the agent into mush.
User request
Cancel order #4812, refund the card on file, and update the dashboard row.
“The winning agent product is not just embedded. It is configured once, shared intentionally, budgeted per user, and visibly trustworthy in every surface where it runs.”
Emcy product principle
Related reads
Docs and architecture guides
Get started
Start with the agent. Then decide where it should live.
Run the agent internally, share it with customers, or embed it directly in your product. Add Gateway when attached servers need OAuth. Use Host when you want Emcy to run the runtime too.