Give your agent production truth.
Hosted MCP server with OAuth, 44 tools, and scoped URLs. Your coding agent uses these tools to run the five-skill playbook.
Pick your client.
Same URL everywhere. Your client signs in with OAuth the first time, so there’s no key to paste.
- 1Add it in chat (works now)
Ask your Bot: "Add a custom MCP server named Talaria at https://api.newtalaria.com/mcp". Confirm the URL, then Authorize and sign in to Talaria in the browser.
- 2Or search Connect apps
Once Talaria appears in Connect apps search, open Connect apps (tap your avatar first on phone), search Talaria, add it, then Authorize.
- 3Check it's installed
Talaria appears under Installed and works in every Bot on your account.
Add a custom MCP server named Talaria at https://api.newtalaria.com/mcpMarketplace listing with five agent skills is submitted and pending. Until it shows in Connect apps search, use the custom Add MCP message — same hosted URL and OAuth. Plugin source: github.com/newtalaria/talaria-agent-plugin. Full playbook: /docs/guides/agent-skills.
- 1Add it in one click
Opens Cursor's install prompt with the server already filled in.
Add to Cursor - 2Sign in
Cursor opens newtalaria.com in your browser. Sign in and choose the organisation.
- 3Ask about production
Try: "What broke after the last deploy in Northline?"
{
"mcpServers": {
"talaria": {
"url": "https://api.newtalaria.com/mcp"
}
}
}Prefer a file? Paste this into ~/.cursor/mcp.json, or .cursor/mcp.json to share it with the repo. For the five-skill playbook, visit cursor.directory/plugins/talaria. Full playbook: /docs/guides/agent-skills.
- 1Install via link (opens VS Code)
One-click install with the server pre-configured.
Install in VS Code - 2Or use the command line
Run the command below, or install the agent plugin from the repository.
- 3Start and sign in
Select Start above the server in mcp.json, then Allow to open the sign-in. The Talaria tools appear in Copilot Chat.
code --add-mcp '{"name":"talaria","type":"http","url":"https://api.newtalaria.com/mcp"}'{
"servers": {
"talaria": {
"type": "http",
"url": "https://api.newtalaria.com/mcp"
}
}
}Not in the VS Code MCP gallery yet. Install with the link or command above.
- 1Install the plugin
Adds the five Talaria skills and connects the MCP server.
- 2Or MCP only
Add the MCP server without the skills.
- 3Use in terminal
Skills and tools are available in Copilot CLI commands.
copilot plugin install newtalaria/talaria-agent-plugin
# Or MCP only:
copilot mcp add --transport http talaria https://api.newtalaria.com/mcpawesome-copilot listing in review.
- 1Add self-hosted marketplace
Registers the plugin marketplace from the repository.
- 2Install the plugin
Installs Talaria from the marketplace you just added.
- 3Sign in and use
Run /mcp in Claude Code, pick talaria and finish the browser sign-in. Reference issues as @talaria:talaria://issue/...
claude plugin marketplace add newtalaria/talaria-agent-plugin
claude plugin install talaria@newtalaria{
"mcpServers": {
"talaria": { "type": "http", "url": "https://api.newtalaria.com/mcp" }
}
}Install from the GitHub repo.
- 1Install from repository
Installs the extension directly from the GitHub repository.
- 2Use with Gemini
The extension tools and skills are available in Gemini CLI commands.
gemini extensions install https://github.com/newtalaria/talaria-agent-pluginNot in the Gemini extensions gallery yet. Install from the GitHub repo.
- 1Add the server
Writes [mcp_servers.talaria] to ~/.codex/config.toml.
- 2Sign in
codex mcp login opens the browser sign-in and stores the token for Codex.
- 3Run with an issue
Ask Codex to read talaria://issue/… with get_error, fix it and call verify_fix.
codex mcp add talaria --url https://api.newtalaria.com/mcp
codex mcp login talaria[mcp_servers.talaria]
url = "https://api.newtalaria.com/mcp"Skills use these tools.
The Talaria plugin brings five agent skills (the playbook) that use the MCP tools below. Available on cursor.directory/plugins/talaria (live) and Grok Bot Connect apps (marketplace pending). Read the full playbook.
Five-skill closed loop
talaria-get-startedtalaria-instrumenttalaria-observe-performancetalaria-triage-issuetalaria-verify-fixHow sign-in works.
Standard OAuth 2.1 with dynamic client registration and PKCE (S256). You approve once per client.
Add the URL
Your client calls https://api.newtalaria.com/mcp and finds Talaria's sign-in through the protected-resource metadata.
Sign in to Talaria
The browser opens newtalaria.com. The client registers itself and uses PKCE, so no secret leaves your machine.
Choose what it can do
Pick the organisation and access: read only, read and write, or with keys. A scoped URL narrows it further.
Revoke any time
Every grant is listed in Settings › Integrations › MCP with its client, scope and last use.
Scopes
Write includes read; keys includes write. Agents ask before any write, and each tool says whether it reads or writes.
No OAuth in your client?
Create a personal token in Settings › Integrations (30 days, 90 days or until revoked) and send it as a bearer token. Never use a tal_live_ ingest key: /mcp rejects them.
44 tools, each marked read or write.
Every tool carries MCP annotations (title, readOnlyHint, destructiveHint), so clients can auto-approve reads and ask before writes.
Docs3
- docs_listList Talaria docsRead
- docs_searchSearch Talaria docsRead
- docs_getRead a Talaria docs pageRead
Connection2
- get_connectionShow MCP connectionRead
- switch_organizationSwitch MCP organizationWrite
Projects6
- get_projectsList projectsRead
- get_projectGet project settingsRead
- update_project_settingsUpdate project settingsWrite
- create_projectCreate projectWrite
- setup_projectSet up project for an SDK installKeys
- send_test_eventSend test eventWrite
Keys1
- create_api_keyCreate ingest API keyKeys
Issues10
- search_errorsSearch issuesRead
- get_errorGet issue with investigation packRead
- update_issue_statusUpdate issue statusWrite
- add_issue_commentComment on issueWrite
- create_github_issueCreate GitHub issueWrite
- link_github_issueLink GitHub issueWrite
- unlink_github_issueUnlink GitHub issueWrite
- get_issue_linksList linked GitHub issuesRead
- verify_fixVerify a fix held after releaseRead
- get_suspect_commitsRank suspect commitsRead
Releases2
- list_releasesList project releasesRead
- get_releaseGet a project releaseRead
Events6
- search_eventsSearch eventsRead
- get_eventGet eventRead
- get_source_mapResolve source-mapped frameRead
- get_replayGet session replay summaryRead
- get_source_contextGet commit-exact source from GitHubRead
- wait_for_eventWait for a matching eventRead
Performance4
- search_tracesSearch tracesRead
- get_traceGet trace waterfallRead
- list_suspect_spansList suspect spansRead
- list_database_queriesList database queriesRead
Analytics2
- search_sessionsSearch sessionsRead
- get_project_statsGet project health statsRead
Flags4
- list_feature_flagsList feature flagsRead
- get_feature_flagGet feature flagRead
- set_feature_flag_offTurn feature flag offWrite
- set_feature_flag_percentSet feature flag rollout percentWrite
Ops3
- list_monitorsList monitorsRead
- list_monitor_runsList monitor runsRead
- get_logsRead logsRead
AI1
- list_agent_runsList recent agent runsRead
Give each agent only what it needs.
Scope by organisation or project, drop the write tools, or pick toolsets. Combine them freely.
https://api.newtalaria.com/mcpEvery organisation you belong to. Use switch_organization to move.
https://api.newtalaria.com/mcp/talaria-labsOne organisation. Tokens made here can't be used on other orgs.
https://api.newtalaria.com/mcp/talaria-labs/northlineOne project. Discovery tools are hidden; the agent starts in Northline.
https://api.newtalaria.com/mcp?readonly=trueOnly read tools are listed. Works with any of the URLs above.
https://api.newtalaria.com/mcp?toolsets=issues,performanceOnly these groups, to keep the agent's context small.
Read only for review agents
Give a code-review bot ?readonly=true. It can read errors, traces and replays but can’t resolve, comment or change flags.
Bound to the URL
A grant made on a scoped URL only works on that URL. Widening the scope means signing in again, so a leaked token can’t reach other projects.
Questions about MCP
Do I need an API key?
No. Clients that support OAuth sign in through the browser. For clients that don't, create a personal token in Settings › Integrations. Ingest keys (tal_live_…) never work on /mcp.
Can the agent change things in production?
Only with write access, and only through tools marked Write: resolving an issue, commenting, changing project settings or a feature flag. Clients ask before running them. Use ?readonly=true to remove them.
Which data does the agent see?
What your Talaria user can see in the organisation or project you chose. Scrubbed fields stay [Filtered]; replays come back as summaries, not recordings.
How do agents fix and verify an issue?
The agent reads the issue with get_error, makes the fix, and — once the release ships — calls verify_fix to confirm the fix held. Dashboard AI is a separate assistant; it reads what you already stored, while MCP lets the agent investigate and verify.
Which clients work?
Any MCP client that speaks Streamable HTTP: Cursor, Claude Code, Codex, VS Code, Grok Bot, Windsurf and Gemini CLI. The docs have a page for each.
How do I revoke access?
Settings › Integrations › MCP lists each grant with its client, scope and last use. Revoke one and that client must sign in again.
Connect your agent, then watch it fix and verify.
Developer is $5 per month. Add Talaria to your client, sign in, and ask what broke.