# emisar — MCP server card for AI agents

Let AI operate servers without SSH. Choose actions, approve risky changes, and audit every step.

Registry name: `dev.emisar/emisar` · category: Other · version 0.60.0 · updated 2026-11-18

## Connect
- Remote (Claude Code): `claude mcp add --transport http emisar https://emisar.dev/api/mcp/rpc`

## Facts
- Repository: https://github.com/andrewdryga/emisar (449 stars)
- Website: https://emisar.dev

Before you connect a server: read its repository, and give it only the permissions and keys it needs.

## Links
- AgentGid page: https://agentgid.com/mcp/servers/dev-emisar-emisar/


## emisar: let your AI work on servers with approvals

emisar lets an AI assistant check and fix your servers without handing it full remote access. It can only pick from a list of prepared actions, like reading logs or restarting a service. Your rules decide what runs, what waits for a person to approve, and what is refused, and every step is logged. It needs an emisar account and a small program on each server.

### Where to use it
- Look into an incident on live servers: When something breaks, your assistant reads logs, service status and disk use across your servers and comes back with evidence and a likely cause.
- Restarts that wait for your yes: Routine checks run on their own, while restarts and deployments pause until a person approves them in the emisar console.
- Find and run a prepared action: The assistant searches the action list by task, reads the exact inputs one action takes, then runs it on the servers you name.
- Check what your own deploy did: After a release misbehaves, the assistant looks at the running result, finds its own mistake and proposes a fix for you to review.
- Turn a fix into a reviewed runbook: The assistant writes a repeatable list of steps as a draft; a person reviews and publishes it before anyone can run it.
- See what the assistant did last night: Review every run the assistant started, with who asked, the reason, the policy decision and the result.

### How to start
- Create an emisar account, then in the console choose Connect a runner and run the generated install command on a Linux server with systemd and sudo.

```
curl -fsSL https://emisar.dev/install.sh \
  | sudo EMISAR_ENROLLMENT_KEY=emkey-enroll-... bash
```

- Check the server is online in the console and run linux.uptime with a reason; the run should appear in the audit trail.
- Open AI agents in the console and connect your app. Claude.ai and ChatGPT sign in directly; local apps use the emisar-mcp bridge.

### License
The server program, bridge and action packs are open source under Apache 2.0. The control plane is under the Business Source License, and the hosted service is free for 3 servers and paid per server above that.

### Good to know
- The program on each server runs allowed actions with its own system permissions and does not sandbox them, so an approved destructive action really happens.
- An optional shell pack gives full command access for emergencies; it is denied by default and should never be installed on production servers.
- The free plan covers 3 servers, 1 user and 7 days of audit history; Team costs $20 per server a month.
- Each server needs Linux with systemd, sudo, and outgoing access to emisar.dev and a few signing hosts.
- The access key acts as you and reaches only the servers and actions your own access allows; it can be revoked at any time.

Checked 2026-11-10 against: https://github.com/AndrewDryga/emisar, https://github.com/AndrewDryga/emisar/blob/31f4c55b3f7513621e09e3f4468ff29dfdc6e123/README.md, https://github.com/AndrewDryga/emisar/blob/31f4c55b3f7513621e09e3f4468ff29dfdc6e123/LICENSE.md, https://github.com/AndrewDryga/emisar/blob/31f4c55b3f7513621e09e3f4468ff29dfdc6e123/server.json, https://github.com/AndrewDryga/emisar/blob/31f4c55b3f7513621e09e3f4468ff29dfdc6e123/mcp/README.md, https://github.com/AndrewDryga/emisar/blob/31f4c55b3f7513621e09e3f4468ff29dfdc6e123/dist/cursor-plugin/mcp.json, https://github.com/AndrewDryga/emisar/blob/31f4c55b3f7513621e09e3f4468ff29dfdc6e123/.agent/kb/specs/mcp-api.md, https://github.com/AndrewDryga/emisar/blob/31f4c55b3f7513621e09e3f4468ff29dfdc6e123/portal/apps/emisar_web/priv/mcp/api-schemas.json, https://emisar.dev/, https://emisar.dev/pricing, https://emisar.dev/privacy, https://emisar.dev/terms, https://emisar.dev/docs/connect-cli-agent, https://emisar.dev/docs/quickstart

### Mistakes to avoid
- Don't: Calling run_action again after its reply was lost → Do: Recover it with get_operation and the same key; never repeat the change (A lost reply does not undo work the control plane already accepted.)
- Don't: Calling run_action straight from a find_actions result → Do: Call get_action first for the exact inputs, risk and side effects, then run_action (The selected action may be destructive and its inputs are strict.)
- Don't: Running find_actions once per server → Do: One find_actions query covers every server in scope (Repeated searches only waste calls.)
- Don't: Committing .mcp.json with EMISAR_API_KEY to the project → Do: Keep the key in user-level or secret config; one key per client (The key acts as you on your servers.)
- Don't: Retrying the same call after invalid_args → Do: Fix the exact paths the error names, then call again (The error lists what to correct; the same call fails again.)

### Questions
- Can the assistant run any command it wants? No. It chooses from declared actions with typed inputs; the server program rejects unknown actions and inputs.
- Who approves risky actions? A person, in the emisar console. Approvals are never decided inside the assistant's chat.
- Does my server need an open port? No. The program on the server only connects outward over an encrypted connection and listens for nothing.
- Is my data used to train AI? emisar says it does not use your data to train AI models.
- Is it free? Free for 3 servers and 1 user with no card; Team is $20 per server a month.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
