Dike's four core MCP tools (dike_search, dike_reason, dike_cite, dike_graph_traverse) are engineered to compose seamlessly inside autonomous agent loops.

---

Pattern 1: Exploratory Research & Hydration

Use this pattern when your agent needs to inspect full statutory provisions before deciding on a legal strategy.

1. Step 1 (dike_search): Send a natural language query with top_k: 5 to identify candidate statutes and court judgements.
2. Step 2 (Filter & Select): The agent inspects the 800-character snippets and selects the most relevant canonical_id.
3. Step 3 (GET /v1/document/{id}): Hydrate the complete document text or article list to examine full legal definitions and exclusions.

---

Pattern 2: Currency & Legislative History Auditing

Never rely on an indexed statute without verifying whether subsequent legislation has amended or repealed its articles.

1. Call dike_graph_traverse with direction: "in" and edge_types: ["amends", "repeals"].
2. Inspect the connected nodes. If an amending statute exists (e.g. eg-leg-2021-45), the agent knows the primary statute has undergone legislative revision.

---

Pattern 3: Direct Grounded Q&A vs. Custom Synthesis

When should an agent call dike_reason versus performing its own synthesis from dike_search?

---

Pattern 4: Rate-Limit & Token-Budget Optimization

In multi-agent swarms or complex tree-of-thought workflows, calling dike_reason at every intermediate sub-step will quickly exhaust rate limit quotas.

Best Practice

• Use dike_search and dike_graph_traverse for exploratory branch evaluation, document filtering, and consistency checks.
• Reserve one single dike_reason invocation for the final user-facing response synthesis.

---

Next Steps

• See these patterns applied in production in Case Studies.
• Review the Grounded Q&A Walkthrough for frontend UI patterns.
• Explore the high-performance gRPC Mirror.