Developer SDKs · Zero External Process Spawning v0.23.0 Integrators Guide

Stop making your AI model sit through a 30-minute monologue

Raw MCP forces models to read giant schema encyclopedias before fetching a single file. Warmplane is the polite control plane that serves a compact menu first—decoupling tool execution from application code.

Try Savings Calculator Explore Integrator APIs →
95.8%
Copilot MCP Token Savings
58.2%
Filesystem Workload Savings
56%
Faster Tool Search (v0.14.0)
01 — Protocol Design Analogy

The House-Guest Lesson

Imagine trapping a guest in your hallway to detail every spice, appliance, and Wi-Fi router firmware history before asking if they want coffee.

Raw MCP (Unfiltered)
The Monologue Host
Injects 200+ raw JSON tool schemas directly into every prompt turn. Causes severe context bloat, slow time-to-first-token, high API bills, and random hallucinations into irrelevant tool definitions.
~38,500 Tokens / Prompt Turn
Warmplane (Control Plane)
The Polite Host
Exposes a fast capability menu (/v1/capabilities/search). Model picks what it needs, Warmplane fetches details on demand and executes safely within policy boundaries.
~1,600 Tokens / Prompt Turn
02 — Dynamic Estimator

Interactive Context Bloat Calculator

See how much context window tax you save per prompt turn as your catalog of integrated MCP tools grows.

Registered MCP Capabilities (Tools): 85 tools
25,500 tok
Raw MCP Prompt Monologue Overhead
1,140 tok (~95.5% Saved)
Warmplane Compact Facade
03 — Developer Architecture

Built for Application Integrators

Warmplane keeps MCP logic decoupled from your application codebase, user authentication, and agent loop orchestration.

P0 • Search
Hybrid Capability Search
Execute fast BM25 + ONNX vector queries over registered tools with natural language intent matching without exposing raw schemas to the model.
P0 • Catalog
SHA-256 Catalog Versioning
Clients safely cache index descriptions and validate state consistency using HTTP ETag headers and monotonic event feeds.
P0 • Observability
Request Correlation
Attach arbitrary application operation IDs (x-correlation-id) to trace exact Warmplane execution trees in OpenTelemetry.
P1 • Recovery
Idempotency & Retries
Safely replay external tool writes after network hiccups using client-provided Idempotency-Key headers and explicit cancellation endpoints.
P1 • Shaping
Result Shaping & Redaction
Filter tool output payloads with JSONPath allowlists, clip giant responses, and automatically redact sensitive secrets before model insertion.
P2 • Security
Scoped Policy Overlays
Dynamically restrict execution scope per request turn without mutating global system security policies.
04 — API & SDK Reference

Embedding Code Reference

Integrate via RESTful HTTP API, Rust client library, or CLI IPC stream.

Embedding Examples
POST /v1/capabilities/search HTTP/1.1 Host: localhost:9090 Content-Type: application/json X-Correlation-ID: op_9f82a1b73c { "query": "triage production errors in logs", "limit": 5, "tags": ["observability", "read_only"] } /* Response: SHA-256 catalog validated compact facade */ HTTP/1.1 200 OK ETag: "sha256:8f2a938c..." { "catalog_version": "sha256:8f2a938c...", "capabilities": [ { "id": "observability.logs.search", "summary": "Search structured application logs.", "score": 0.94 } ] }
use warmplane_client::WarmplaneClient; #[tokio::main] async fn main() -> Result<(), Box<dyn std::error::Error>> { let client = WarmplaneClient::connect("http://localhost:9090")?; // Search compact catalog without schema injection let results = client.search_capabilities() .query("triage production errors") .limit(5) .send() .await?; println!("Found {} candidate capabilities", results.capabilities.len()); Ok(()) }
# Search capability catalog via CLI JSON stream warmplane search-capabilities "triage production errors" --limit 5 --json # Execute capability with context correlation header warmplane call-capability observability.logs.search \ --params '{"query":"error 500"}' \ --correlation-id "task_8812"
Open Source · MIT License

The cheapest token is
the one never sent

Start cutting AI agent token overhead while adding enterprise policy control.

GitHub Repository ↗ User Guide Read Whitepaper