Agent guides (16)
Connect JetBrains AI Assistant and Junie to Ishara
Connect JetBrains AI Assistant or Junie to Ishara through the mcp-remote bridge, since neither sends request headers itself.
- Server
https://useishara.com/mcp- Where it goes
- Settings → Tools → AI Assistant → Model Context Protocol (MCP), or
.junie/mcp/mcp.json - Transport
- stdio, through
mcp-remote - Authentication
- A bearer token in the
Authorizationheader - Time
- About 4 minutes
Steps
In Ishara, open Agents, name a token under New token (for example “JetBrains (AI Assistant, Junie) on my laptop”) and click Create token. Copy it now: it won’t be shown again.
Add Ishara to JetBrains AI Assistant and Junie, with your token in place of
ishara_…:{ "mcpServers": { "ishara": { "command": "npx", "args": ["-y", "mcp-remote", "https://useishara.com/mcp", "--header", "Authorization:${AUTH_HEADER}", "--transport", "http-only"], "env": { "AUTH_HEADER": "Bearer ishara_…" } } } }AI Assistant: add a server in its MCP settings and paste this. Junie: save it as
.junie/mcp/mcp.jsonin the project, or~/.junie/mcp/mcp.json. Needs Node.js 18 or later.Click Apply. The Status column shows the connection; hover over the icon to see an error.
Ask your agent: “Apply the open Ishara notes.” It lists them, applies each edit in your code, and marks each note done with a reply the reviewer sees on the site.
Good to know
- AI Assistant’s MCP settings have no field for headers, and Junie’s docs say it “does not support security tokens in MCP configs”, so the bridge runs on your computer and adds the token.
Authorization:${AUTH_HEADER}has no space after the colon on purpose: some clients split arguments on spaces.mcp-remoteis a community package, not a JetBrains feature.
What your agent can do
| Tool | What it does |
|---|---|
list_sites | The sites (embeds) in this account, with how many notes are open on each and what agents may see there. |
list_notes | Notes reviewers left on the live sites, newest first. Default: open notes (new and in progress) on every site. Each has an id for get_note and update_note, and a short ref like "N-12". |
get_note | One note in full: the page, the section heading, the CSS selector, the text before and after, the reviewer's words and its history. |
get_note_photo | The photo a reviewer attached (the replacement image, or one that shows the problem), as an image. With which: "screenshot", the screenshot of the page as the reviewer saw it, any element they picked ringed in gold (get_note says whether a note has one). |
update_note | Set a note's status and/or reply. The reviewer sees both in the widget on their site. Use "doing" while working, "done" once the change is committed or deployed, "wontdo" with a reason when it will not be applied. Write the reply in the reviewer's language. |
export_notes | Every open note, grouped by page, as one Markdown document (a checklist to work through). |
Reviewers are people outside your team, so their words reach the agent marked as untrusted data, never as instructions. On sites where anyone can leave a note, set the site to Accepted notes only: agents see a note only after a person moves it to In progress.