diff --git a/cli/main.py b/cli/main.py
new file mode 100644
index 0000000..777ec2e
--- /dev/null
+++ b/cli/main.py
@@ -0,0 +1,201 @@
+"""Marchwarden CLI shim.
+
+Talks to the web researcher MCP server over stdio and pretty-prints
+ResearchResult contracts to the terminal.
+"""
+
+import asyncio
+import json
+import sys
+from typing import Optional
+
+import click
+from mcp import ClientSession, StdioServerParameters
+from mcp.client.stdio import stdio_client
+from rich.console import Console
+from rich.panel import Panel
+from rich.table import Table
+from rich.text import Text
+
+from researchers.web.models import ResearchResult
+
+
+# ---------------------------------------------------------------------------
+# MCP client
+# ---------------------------------------------------------------------------
+
+
+async def call_research_tool(
+    question: str,
+    depth: str,
+    max_iterations: int,
+    token_budget: int,
+) -> ResearchResult:
+    """Spawn the web researcher MCP server and call its `research` tool."""
+    params = StdioServerParameters(
+        command=sys.executable,
+        args=["-m", "researchers.web.server"],
+    )
+    async with stdio_client(params) as (read, write):
+        async with ClientSession(read, write) as session:
+            await session.initialize()
+            result = await session.call_tool(
+                "research",
+                arguments={
+                    "question": question,
+                    "depth": depth,
+                    "max_iterations": max_iterations,
+                    "token_budget": token_budget,
+                },
+            )
+            # FastMCP returns the tool's string return as a TextContent block.
+            payload = result.content[0].text
+            return ResearchResult.model_validate_json(payload)
+
+
+# ---------------------------------------------------------------------------
+# Pretty printing
+# ---------------------------------------------------------------------------
+
+
+def render_result(result: ResearchResult, console: Console) -> None:
+    """Render a ResearchResult to the console using rich."""
+    # Answer
+    console.print(
+        Panel(
+            result.answer,
+            title="[bold cyan]Answer[/bold cyan]",
+            border_style="cyan",
+        )
+    )
+
+    # Citations
+    if result.citations:
+        table = Table(title="Citations", show_lines=True, expand=True)
+        table.add_column("#", style="dim", width=3)
+        table.add_column("Title / Locator", overflow="fold")
+        table.add_column("Excerpt", overflow="fold")
+        table.add_column("Conf", justify="right", width=5)
+        for i, c in enumerate(result.citations, 1):
+            header = f"[bold]{c.title or c.locator}[/bold]\n[dim]{c.locator}[/dim]"
+            table.add_row(str(i), header, c.raw_excerpt, f"{c.confidence:.2f}")
+        console.print(table)
+    else:
+        console.print("[dim]No citations.[/dim]")
+
+    # Gaps grouped by category
+    if result.gaps:
+        gap_table = Table(title="Gaps", show_lines=True, expand=True)
+        gap_table.add_column("Category", style="yellow")
+        gap_table.add_column("Topic")
+        gap_table.add_column("Detail", overflow="fold")
+        for g in result.gaps:
+            gap_table.add_row(g.category.value, g.topic, g.detail)
+        console.print(gap_table)
+
+    # Discovery events
+    if result.discovery_events:
+        de_table = Table(title="Discovery Events", show_lines=True, expand=True)
+        de_table.add_column("Type", style="magenta")
+        de_table.add_column("Suggested Researcher")
+        de_table.add_column("Query", overflow="fold")
+        de_table.add_column("Reason", overflow="fold")
+        for d in result.discovery_events:
+            de_table.add_row(
+                d.type, d.suggested_researcher or "-", d.query, d.reason
+            )
+        console.print(de_table)
+
+    # Open questions
+    if result.open_questions:
+        oq_table = Table(title="Open Questions", show_lines=True, expand=True)
+        oq_table.add_column("Priority", style="green")
+        oq_table.add_column("Question", overflow="fold")
+        oq_table.add_column("Context", overflow="fold")
+        for q in result.open_questions:
+            oq_table.add_row(q.priority, q.question, q.context)
+        console.print(oq_table)
+
+    # Confidence + factors
+    cf = result.confidence_factors
+    conf_text = Text()
+    conf_text.append(f"Overall: {result.confidence:.2f}\n", style="bold")
+    conf_text.append(f"Corroborating sources: {cf.num_corroborating_sources}\n")
+    conf_text.append(f"Source authority: {cf.source_authority}\n")
+    conf_text.append(f"Contradiction detected: {cf.contradiction_detected}\n")
+    conf_text.append(f"Query specificity match: {cf.query_specificity_match:.2f}\n")
+    conf_text.append(f"Budget exhausted: {cf.budget_exhausted}\n")
+    conf_text.append(f"Recency: {cf.recency or 'unknown'}")
+    console.print(Panel(conf_text, title="Confidence", border_style="green"))
+
+    # Cost
+    cm = result.cost_metadata
+    cost_text = Text()
+    cost_text.append(f"Tokens: {cm.tokens_used}\n")
+    cost_text.append(f"Iterations: {cm.iterations_run}\n")
+    cost_text.append(f"Wall time: {cm.wall_time_sec:.2f}s\n")
+    cost_text.append(f"Model: {cm.model_id}")
+    console.print(Panel(cost_text, title="Cost", border_style="blue"))
+
+    # Trace footer
+    console.print(f"\n[dim]trace_id: {result.trace_id}[/dim]")
+
+
+# ---------------------------------------------------------------------------
+# Click app
+# ---------------------------------------------------------------------------
+
+
+@click.group()
+def cli() -> None:
+    """Marchwarden — agentic research CLI."""
+
+
+@cli.command()
+@click.argument("question")
+@click.option(
+    "--depth",
+    type=click.Choice(["shallow", "balanced", "deep"]),
+    default="balanced",
+    show_default=True,
+)
+@click.option(
+    "--budget",
+    "token_budget",
+    type=int,
+    default=20_000,
+    show_default=True,
+    help="Token budget for the research loop.",
+)
+@click.option(
+    "--max-iterations",
+    type=int,
+    default=5,
+    show_default=True,
+)
+def ask(
+    question: str,
+    depth: str,
+    token_budget: int,
+    max_iterations: int,
+) -> None:
+    """Ask the web researcher a QUESTION."""
+    console = Console()
+    console.print(f"[dim]Researching:[/dim] {question}\n")
+    try:
+        result = asyncio.run(
+            call_research_tool(
+                question=question,
+                depth=depth,
+                max_iterations=max_iterations,
+                token_budget=token_budget,
+            )
+        )
+    except Exception as e:
+        console.print(f"[bold red]Error:[/bold red] {e}")
+        sys.exit(1)
+    render_result(result, console)
+
+
+if __name__ == "__main__":
+    cli()
diff --git a/pyproject.toml b/pyproject.toml
index 62f9ab4..e463baf 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -19,6 +19,7 @@ dependencies = [
     "tavily-python>=0.3.0",
     "httpx>=0.24.0",
     "click>=8.0",
+    "rich>=13.0",
 ]
 
 [project.optional-dependencies]
diff --git a/tests/test_cli.py b/tests/test_cli.py
new file mode 100644
index 0000000..189faa8
--- /dev/null
+++ b/tests/test_cli.py
@@ -0,0 +1,132 @@
+"""Tests for the marchwarden CLI."""
+
+from unittest.mock import patch
+
+from click.testing import CliRunner
+
+from cli.main import cli, render_result
+from researchers.web.models import (
+    Citation,
+    ConfidenceFactors,
+    CostMetadata,
+    DiscoveryEvent,
+    Gap,
+    GapCategory,
+    OpenQuestion,
+    ResearchResult,
+)
+from rich.console import Console
+
+
+def _fixture_result() -> ResearchResult:
+    return ResearchResult(
+        answer="Tomatoes, peppers, squash, and beans grow well in Utah.",
+        citations=[
+            Citation(
+                source="web",
+                locator="https://extension.usu.edu/yard-and-garden",
+                title="USU Extension — Yard and Garden",
+                snippet="USU recommends warm-season crops for Utah's climate.",
+                raw_excerpt="Tomatoes, peppers, and squash thrive in Utah summers.",
+                confidence=0.9,
+            ),
+        ],
+        gaps=[
+            Gap(
+                topic="Microclimate variation",
+                category=GapCategory.SCOPE_EXCEEDED,
+                detail="Did not investigate elevation-specific recommendations.",
+            ),
+        ],
+        discovery_events=[
+            DiscoveryEvent(
+                type="related_research",
+                suggested_researcher="docs",
+                query="Utah USDA hardiness zones",
+                reason="Zone-specific guidance would improve answer.",
+            ),
+        ],
+        open_questions=[
+            OpenQuestion(
+                question="What are the best cool-season crops?",
+                context="Answer focused on warm-season crops.",
+                priority="medium",
+            ),
+        ],
+        confidence=0.82,
+        confidence_factors=ConfidenceFactors(
+            num_corroborating_sources=3,
+            source_authority="high",
+            contradiction_detected=False,
+            query_specificity_match=0.85,
+            budget_exhausted=False,
+            recency="current",
+        ),
+        cost_metadata=CostMetadata(
+            tokens_used=4321,
+            iterations_run=3,
+            wall_time_sec=12.5,
+            budget_exhausted=False,
+            model_id="claude-sonnet-4-6",
+        ),
+        trace_id="trace-abc-123",
+    )
+
+
+class TestRenderResult:
+    def test_renders_all_sections(self):
+        console = Console(record=True, width=120)
+        render_result(_fixture_result(), console)
+        out = console.export_text()
+        assert "Tomatoes" in out
+        assert "USU Extension" in out
+        assert "scope_exceeded" in out
+        assert "related_research" in out
+        assert "cool-season" in out
+        assert "Confidence" in out
+        assert "claude-sonnet-4-6" in out
+        assert "trace-abc-123" in out
+
+
+class TestAskCommand:
+    def test_ask_invokes_mcp_and_renders(self):
+        runner = CliRunner()
+        fixture = _fixture_result()
+
+        async def fake_call(question, depth, max_iterations, token_budget):
+            assert question == "What grows in Utah?"
+            assert depth == "shallow"
+            assert max_iterations == 2
+            assert token_budget == 5000
+            return fixture
+
+        with patch("cli.main.call_research_tool", side_effect=fake_call):
+            result = runner.invoke(
+                cli,
+                [
+                    "ask",
+                    "What grows in Utah?",
+                    "--depth",
+                    "shallow",
+                    "--max-iterations",
+                    "2",
+                    "--budget",
+                    "5000",
+                ],
+            )
+
+        assert result.exit_code == 0, result.output
+        assert "Tomatoes" in result.output
+        assert "trace-abc-123" in result.output
+
+    def test_ask_handles_error(self):
+        runner = CliRunner()
+
+        async def boom(**kwargs):
+            raise RuntimeError("mcp went sideways")
+
+        with patch("cli.main.call_research_tool", side_effect=boom):
+            result = runner.invoke(cli, ["ask", "anything"])
+
+        assert result.exit_code == 1
+        assert "mcp went sideways" in result.output