Agents and MCP
Connect coding agents to Talaria: skills are the playbook, MCP is the tool belt. Get started, instrument, observe, triage, verify.
The Talaria plugin for Cursor and Grok Bot includes five closed-loop skills that guide agents from initial connection through fix verification. These skills wrap the hosted MCP tools and reference the detailed playbooks in this docs repo.
Plugin: github.com/newtalaria/talaria-agent-plugin
Listing: cursor.directory/plugins/talaria
Closed-loop workflow
talaria-get-started → talaria-instrument → talaria-observe-performance → talaria-triage-issue → talaria-verify-fixEach skill focuses on one phase, hands off to the next when appropriate, and constrains writes to prevent silent project creation or status changes.
Skills → playbooks map
| Skill | Purpose | Core tools | Existing playbook |
|---|---|---|---|
talaria-get-started |
Connect MCP, confirm org/project, first smoke read | get_connection, list_projects, get_doc, list_docs, and search_docs, search_issues |
Orientation only — no dedicated playbook |
talaria-instrument |
SDK install, create project/key, send test event | setup_project, send_test_event, search_events, get_doc |
Add Talaria with an agent |
talaria-observe-performance |
Health, traces, spans, queries, sessions, monitors | get_project_stats, search_traces, list_suspect_spans, list_database_queries |
Read-only stats — extends playbook tools |
talaria-triage-issue |
Investigate one issue: pack → trace/replay → optional GitHub | get_issue, get_trace, get_replay, list_suspect_commits, create_github_issue, update_issue_status |
Triage issues with an agent |
talaria-verify-fix |
After release: verdict + optional status update | verify_fix, list_releases, get_release, wait_for_event, update_issue_status |
Extends triage with release verification — see verify_fix tool in MCP |
Where agents get these skills
From the Talaria plugin
Install the Talaria plugin. The plugin includes:
mcp.jsonpointing athttps://api.newtalaria.com/mcp- Five skill
SKILL.mdfiles inskills/with YAML frontmatter
Agents that load the plugin automatically discover and apply these skills when relevant.
From Markdown playbooks via get_doc
Agents that do not have the plugin installed can still follow the same workflows by reading the playbooks via MCP get_doc:
- Install:
get_docpathagents/add-talaria - Triage:
get_docpathagents/triage-issues - Configuration:
get_docpathget-started/project-settings - SDK reference:
get_docpathsdk/<id>/README(e.g.sdk/flutter/README)
The playbooks are the detailed how-to; the skills are workflow guardrails that enforce read-only smoke, ask before create, and hand off appropriately.
Tool scopes
| Scope | Includes |
|---|---|
mcp:read |
get_connection, list_projects, get_project, search_issues, get_issue, search_events, get_event, search_traces, get_trace, get_replay, list_suspect_commits, get_source_context, get_source_map, list_suspect_spans, list_database_queries, search_sessions, get_project_stats, feature flag reads, list_releases, get_release, wait_for_event, verify_fix |
mcp:write |
All read tools plus update_project_settings, create_project, send_test_event, update_issue_status, add_issue_comment, create_github_issue, link_github_issue, unlink_github_issue, feature flag writes |
mcp:keys |
All write tools plus create_api_key, setup_project |
Install scopes are mcp:read mcp:write mcp:keys. Skills ask before using write tools.
Skill usage examples
First-time setup
User: "Add Talaria to this Flutter project"
talaria-get-started→ authenticate, confirm orgtalaria-instrument→ detect Flutter,get_docSDK hub,setup_projectwithconfirmed: true, wiretalaria_flutter,send_test_event,search_eventswithenvironment: development- Hand off to
talaria-observe-performanceortalaria-triage-issuewhen ready
Production investigation
User: "What's failing in production?" or paste a Talaria issue URL
talaria-get-started→ confirm connection (may skip if already connected)talaria-triage-issue→get_issuewithissueId, read pack (latest instance, breadcrumbs, release, suspects), optionalget_trace/get_replay, state hypothesis, optional GitHub create/link
Post-release verification
User: "Did version 2.1.0 fix issue NORTH-4F2?"
talaria-verify-fix→list_releases,get_release,verify_fixwithissueId+release, report verdict (verified,regressed,pending_adoption, etc.), optionalupdate_issue_statustoresolvedafter user approval
Skills vs playbooks
| Source | Best for | Provides |
|---|---|---|
| Skills (plugin) | Agents with plugin support (Cursor, Grok Bot) | Workflow guardrails, read-only checks, ask-before-write, hand-offs |
Playbooks (get_doc) |
Any agent via MCP get_doc, list_docs, and search_docs tools |
Detailed how-to, SDK snippets, conversation scripts, must-not rules |
Skills reference playbooks when needed (get_doc path agents/add-talaria). Playbooks are the canonical source of truth for SDK wiring; skills enforce the workflow.
Related
- MCP overview — tool catalogue, scopes, auth
- Add Talaria with an agent — install playbook
- Triage issues with an agent — investigation playbook
- Cursor MCP setup
- Grok Bot MCP setup
- Talaria plugin repo