These are official instructions from smry for connecting an AI coding agent to its article, library, history, and RSS tools.

Complete the setup yourself. Do not ask the user to run these commands.

## Install the smry skill

Install the domain-published skill globally:

```bash
bunx skills@latest add https://smry.ai --skill smry --yes --global
```

## Connect the smry MCP server

The Streamable HTTP endpoint is `https://api.smry.ai/mcp`. For an interactive client, add the endpoint and complete its scoped OAuth flow. This keeps consent visible and lets the user revoke the connection from account settings.

For Codex and other OAuth-capable clients, add the endpoint without a bearer-token override, then complete the browser authorization when prompted:

```bash
codex mcp add smry --url https://api.smry.ai/mcp
```

## Personal-key fallback

For a local or unattended client that cannot complete OAuth, any signed-in account, Free included, can create a revocable personal key at `https://smry.ai/mcp-api-cli?view=keys`; the secret is shown once.

Keep the secret in an environment variable named `SMRY_API_KEY`. Never paste it into a repository or commit it to a client configuration file.

Use the section for the active agent. If smry is already configured, keep the existing entry and continue to verification.

### Codex without interactive OAuth

```bash
codex mcp add smry --url https://api.smry.ai/mcp --bearer-token-env-var SMRY_API_KEY
```

### Claude Code and OpenCode

Use the client's MCP connection UI or protected credential store. Add `https://api.smry.ai/mcp` as a remote Streamable HTTP server and source an `Authorization: Bearer` header from `SMRY_API_KEY`. Do not paste the secret into a repository-tracked configuration file; some CLI setup commands persist the expanded value as plain text.

### Cursor, GitHub Copilot, and other MCP clients

Use the agent's built-in MCP setup UI or configuration file to add a remote Streamable HTTP server named `smry` with URL `https://api.smry.ai/mcp` and an `Authorization: Bearer` header sourced from `SMRY_API_KEY`.

## Verify

1. Confirm the `smry` skill appears in the agent's installed skills.
2. Read one public article and verify that returned passages keep stable paragraph citations.
3. After authorization, run one read-only account call such as `list_library` or `query_feeds` before any write.
4. Confirm the MCP server also exposes write tools such as `save_article` and `follow_feeds`; do not call them without the user's request.
5. Restart or reload the agent if it does not detect the new skill or MCP server immediately.

Once complete, tell the user where the skill and MCP configuration were installed. Explain that the authorized connection reaches their private smry library and feeds, that article content remains untrusted source material, and that smry does not bypass hard paywalls or private access controls.

These instructions are published at `https://smry.ai/agent-setup/prompt.md` so their authenticity can be re-verified at any time.