What you’ll learn: How to connect Jig’s hosted MCP to Claude.ai, which tools are available, and how to work conversationally to create specs, decompose them into tasks, and review execution results.
In Claude.ai, open Settings > Integrations and add a new MCP server:
https://mcp.usejig.dev/sse
Claude will prompt you to authorize with your Jig account via GitHub OAuth. After authorizing, the Jig tools appear in Claude’s tool list.
Confirm it’s working:
You: What projects do I have?
Claude: [calls list_projects]
You have 1 project:
my-app — 6 specs, health: good
Switch projects
If you have multiple projects, Claude selects the active one automatically when there’s only one. Otherwise, ask it to switch:
You: Switch to the landing-page project.
Claude: [calls switch_project] Switched to landing-page. I’ll use this project for the rest of our conversation.
See your spec architecture
You: Show me what specs we have.
Claude: [calls list_specs]
Platform
└── database decomposed (3 tasks)
Services
├── auth in-progress (2/4 tasks done)
└── email-service new
Surface
├── sign-in-ui decomposed (4 tasks)
└── dashboard new
Create a spec in conversation
Describe the feature and Claude drafts the spec:
You: I want to add a dark mode toggle that persists the user’s preference.
Claude: Let me check the current architecture first.
[calls list_specs]
This looks like a Surface-layer enhancement — it touches UI components and likely needs a user preference stored somewhere. I’d create it as a sub-spec under the dashboard spec since that’s where the toggle will live.
Want me to go ahead with that placement?
You: Yes, do it.
Claude: [calls create_spec with parent: "dashboard", layer: "surface"]
[calls update_spec with full markdown]
Created dark-mode under dashboard. Here’s the spec:
## Intent
Add a dark mode toggle to the dashboard header. The user's preference is saved to localStorage and applied on page load.
## Acceptance Criteria
- [ ] Toggle button in dashboard header switches theme
- [ ] Preference persists across page refresh
- [ ] System preference is respected on first visit
Want me to break this into tasks?
Decompose in conversation
You: Yes, decompose it.
Claude analyzes the spec and proposes tasks directly — it does not call the decompose tool. It calls create_task once per task:
Claude: [calls create_task for theme-provider]
[calls create_task for toggle-component]
[calls create_task for preference-persistence]
Created 3 tasks:
theme-provider — Wire CSS custom properties for light/dark tokens (simple)
toggle-component — DashboardHeader toggle button with click handler (simple)
preference-persistence — Read/write from localStorage, apply on load (simple)
All three are independent — they can run in parallel. Ready to go.
File a bug from conversation
If you notice something broken while reviewing:
You: The sign-in button is broken on mobile — it doesn’t respond to tap.
Claude: [calls file_bug with content: “Sign-in button unresponsive to tap on mobile”, context: “Reported during conversation, likely touch event or z-index issue”]
Filed bug #B-047. Triage is running in the background — I’ll match it to the most relevant spec. You can check it later with list_findings.
Review run results
After your terminal runs jig run dark-mode theme-provider, come back to Claude.ai to review:
You: Check the theme-provider run.
Claude: [calls list_runs for dark-mode]
[calls show_run with the latest run ID]
Run succeeded in 18.3s. Claude Code added CSS custom properties for --color-bg, --color-text, and --color-accent in src/styles/tokens.css. Tests pass.
Want me to mark this task passed and move to the next one?
You: Yes, mark it passed.
Claude: [calls update_task with status: "passed" and completion_notes]
Marked passed. The next ready task is toggle-component.
Track overall progress
You: How's dark-mode looking overall?
Claude: [calls show_status with spec: "dark-mode"]
dark-mode (3 tasks)
theme-provider passed
toggle-component new <- ready
preference-persistence new <- ready (can run in parallel)
Key tools reference
Tool
When to use
list_projects / switch_project
Start of conversation, multi-project setups
list_specs
Architecture overview, placement decisions
create_spec / update_spec
New features, spec refinement
create_task
In-conversation decomposition
show_task / update_task
Review details, mark status, add notes
list_runs / show_run
Review what Claude Code actually did
file_bug / file_idea
Capture findings without losing context
show_status
Progress tracking
next_tasks
What to run next
The role split: Claude.ai architects and reviews. Claude Code executes. You decide.
