Real workflows
Use cases that prove MCP delivers real work
Persona-organized demos show exactly what runs locally, when approvals trigger, and which artifacts you get back.
60-second demo: Plan → Execute → Verify
Watch MCP summarize meeting notes, generate an action list, and route approvals before sharing results.
Autonomy budget: 5 minutes • Policy: constrained • Outbound: off
Download the sample project:
git clone https://github.com/completeassistant/demoGoal
Plan → Execute → Verify loop orchestrated locally.
Tools
docs.search_docs, code.search, llm-local.generate, eval.run_unit_tests
Policy
Constrained mode with approvals for writer tools and outbound.
Eval results: rubric pass, non-echo completion confirmed, logs signed (planned).
Choose the path that matches your needs
Each scenario breaks down the exact steps the assistant performs, what stays offline, and where approvals sit.
Individual Productivity
Talk to us about individual productivityAutomate personal admin while keeping everything local and reviewable.
Summarize weekly notes into an action list
Upload markdown notes via docs.index_blob (gated)
Orchestrator plans docs.search_docs → llm-local.generate
eval.score_qa checks structure before creating tasks
All steps run offline (docs + llm-local + eval).
Approval required before writing tasks to your system.
Prioritized task list, summary markdown, audit log entry.
10 minutes saved every week
Draft weekly update email; queue for review
Vector recall of last week's highlights
llm-local.generate builds draft
comms.send_message (gated) awaits approval
Drafting and recall stay local; sending is optional.
Manual approval before outbound email.
Email draft in markdown, status timeline.
Saves ~30 minutes per report
Research & Analysis
Talk to us about research & analysisCollect sources, extract claims, and keep citations auditable.
Collect sources, extract key claims, cross-reference quotes
browser.goto restricted domains (allowlisted)
browser.query extracts relevant sections
llm-local.generate builds claim table
eval.score_qa checks citation completeness
Parsing and synthesis stay on-device; browsing limited to allowlist.
Outbound browsing approved per domain list.
Citation table (CSV), evidence log, stored snapshots.
Hours saved per research sprint
Developer Workflow
Talk to us about developer workflowShip safer patches with gated writes and evaluation hooks.
Add input validation and unit tests; open diff preview
code.search identifies target module
code.propose_patch (gated) drafts changes
eval.run_unit_tests executes stub tests
Code search, diff creation, and eval tests run locally.
Required before patch apply or PR submission.
Diff preview, test report, approval transcript.
Saves 45 minutes per change
Team Operations
Talk to us about team operationsCoordinate projects with shared visibility and policy controls.
Generate project plan with milestones and owners
vector.search retrieves project history
llm-local.generate drafts plan
docs.write_doc (gated) publishes to shared space
comms.draft_message prepares Slack summary
Planning and drafting remain local; publishing gated.
Doc write and Slack post require approval.
Project plan, owner matrix, Slack draft, audit log.
Saves half a day of coordination
Ready to run these workflows yourself?
Download MCP, import the sample tasks, and explore how approvals, budgets, and evaluations keep every run safe.