Verified Burst
MCP stdio server, one tool buy_verified_burst — x402 pay-only-if-verified inference.
README
verified-burst broker
Sell verified inference bursts on Cerebras. An agent hits a hard / irreversible / low-confidence decision and buys more thinking: escalate to fast silicon → best-of-N → verify → pay per burst over x402, charged only if the answer passes. Budget-capped per agent.
BYOK passthrough: the customer's tokens are billed to their own Cerebras key — we never mark up tokens. We charge a small service fee for routing + verification + the burst guarantee. (Legal-clean: selling our application, not reselling their API.)
The wedge — one line for an agent builder
{ "mcpServers": { "verified-burst": {
"command": "python3", "args": ["/root/inference-burst/mcp_server.py"] } } }
Their agent gains one tool, buy_verified_burst(request, strategy, n, verifier, answer_key),
that returns a verified answer + receipt and bills per burst.
Files
| file | role |
|---|---|
provider.py |
BYOK Cerebras call (OpenAI-compatible). Key never stored/logged. |
burst.py |
verified-burst core: best-of-N + self-consistency / deterministic-check / judge verifiers. passed gates settlement. |
pricing.py |
service-fee quote (not a token markup). |
x402_gate.py |
x402 challenge + facilitator verify/settle. Settle only on passed = pay-only-if-verified. SIM until X402_FACILITATOR_URL set. |
broker.py |
orchestration: quote → authorize → burst → settle-IF-verified, + per-agent budget cap. |
server.py |
HTTP surface (POST /v1/burst, GET /v1/quote, /healthz). |
mcp_server.py |
MCP stdio surface — the one-liner. |
measure.py |
the 1-day BYOK measurement harness (offline + live). |
Run
# offline measurement / shape
python3 measure.py --offline
# live (needs .env with CEREBRAS_API_KEY)
python3 measure.py
# HTTP broker
PORT=8402 python3 server.py
# MCP server is launched by the MCP client via the config above
Config (.env, gitignored)
CEREBRAS_API_KEY=csk-... # required (BYOK)
CEREBRAS_MODEL=gpt-oss-120b # this account: gpt-oss-120b | zai-glm-4.7 (NOT llama)
# x402 (REAL mode — omit for SIM):
X402_FACILITATOR_URL=... # facilitator that verifies/settles payments
X402_PAY_TO=0x... # seller wallet that receives USDC
X402_USDC_ASSET=0x... # USDC contract on the chosen network
X402_NETWORK=base-sepolia
Go-live checklist
- [ ] x402 creds: facilitator URL + seller wallet (
X402_PAY_TO) + USDC asset. Until set, payments run in SIM (flow is real, settlement is stubbed). - [ ] Verify pricing:
gpt-oss-120bper-token rate (current placeholder is UNVERIFIED) and setFEESinpricing.pyto a margin you've checked. - [ ] Pick the verifier per use case:
self_consistency(cheap, default),judge(adversarial, costs an extra call), or a callerdeterministic_check(gold standard). - [ ] Budget cap:
broker.DEFAULT_BUDGET_USD— wire to the AgentsPrice margin governor for real per-agent caps. - [ ] Cerebras gotcha: requests need a browser-ish
User-Agentor Cloudflare returns 403 error 1010 (already handled).
Status (2026-06-15)
Core + both surfaces run live against real Cerebras. Verified→charged, failed→not charged, no-payment→402, budget-cap→refused — all confirmed. x402 in SIM pending creds.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.