double-check-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| GEMINI_API_KEY | No | API key for Gemini provider. | |
| OPENAI_API_KEY | No | API key for OpenAI provider. | |
| DEEPSEEK_API_KEY | No | API key for DeepSeek provider. | |
| ANTHROPIC_API_KEY | No | API key for Anthropic provider. | |
| VERIFIER_PLATFORM_KB_PATH | No | Absolute path to custom knowledge base Markdown file. | |
| VERIFIER_PLATFORM_DATA_DIR | No | Default directory for keys and audit data. | ~/.verifier-platform |
| VERIFIER_PLATFORM_HTTP_HOST | No | HTTP listen address. | 127.0.0.1 |
| VERIFIER_PLATFORM_HTTP_PORT | No | HTTP listen port. | 4200 |
| VERIFIER_PLATFORM_LLM_MODEL | No | Model name for the verification LLM. Defaults per provider. | |
| VERIFIER_PLATFORM_TRACE_DIR | No | Directory for per-run trace files. | |
| VERIFIER_PLATFORM_AUDIT_PATH | No | Path to the audit JSONL file. | |
| VERIFIER_PLATFORM_KB_ENABLED | No | Whether to enable the local knowledge base retrieval. | true |
| VERIFIER_PLATFORM_LLM_API_KEY | No | Generic API key that takes priority over provider-specific keys. | |
| VERIFIER_PLATFORM_LLM_BASE_URL | No | Override built-in API root URL. Required for openai-compatible provider. | |
| VERIFIER_PLATFORM_LLM_PROVIDER | No | The LLM provider: openai, deepseek, anthropic, gemini, or openai-compatible. Defaults to deepseek. | deepseek |
| VERIFIER_PLATFORM_AUDIT_ENABLED | No | Whether to append JSONL audit logs. | true |
| VERIFIER_PLATFORM_HTTP_MCP_PATH | No | MCP URL path for HTTP mode. | /mcp |
| VERIFIER_PLATFORM_SYSTEM_PROMPT | No | Custom system prompt for the verification model. | |
| VERIFIER_PLATFORM_TRACE_ENABLED | No | Whether to save per-run trace files. | true |
| VERIFIER_PLATFORM_KB_MAX_RESULTS | No | Maximum number of knowledge base sections to provide to the verifier (5 or 10). | 5 |
| VERIFIER_PLATFORM_LLM_TIMEOUT_MS | No | Model call timeout in milliseconds. | 120000 |
| VERIFIER_PLATFORM_SIGNING_KEY_ID | No | Fixed Ed25519 public key fingerprint (sha256: + 64 hex). Servers refuse to start/sign if fingerprint does not match. | |
| VERIFIER_PLATFORM_LLM_TEMPERATURE | No | Temperature for the verification model. | 0 |
| VERIFIER_PLATFORM_MAX_INPUT_CHARS | No | Maximum total input characters for system prompt and verification input. | 200000 |
| VERIFIER_PLATFORM_PUBLIC_KEY_PATH | No | Custom path to the Ed25519 public key. | |
| VERIFIER_PLATFORM_CONFIDENCE_FLOOR | No | Confidence threshold for policy.pass. | 0.70 |
| VERIFIER_PLATFORM_LLM_HEADERS_JSON | No | Additional request headers for the LLM as a JSON string. | {} |
| VERIFIER_PLATFORM_PRIVATE_KEY_PATH | No | Custom path to the Ed25519 private key. | |
| VERIFIER_PLATFORM_HTTP_BEARER_TOKEN | No | Optional Bearer token for HTTP authorization. | |
| VERIFIER_PLATFORM_HTTP_ALLOWED_HOSTS | No | Comma-separated list of allowed Host/Origin values. | localhost,127.0.0.1,[::1] |
| VERIFIER_PLATFORM_WEB_SEARCH_ENABLED | No | Whether web search is enabled by default when callers omit webSearch. | false |
| VERIFIER_PLATFORM_LLM_MAX_OUTPUT_TOKENS | No | Maximum output tokens for verification. | 2048 |
| VERIFIER_PLATFORM_WEB_SEARCH_TIMEOUT_MS | No | Timeout for DeepSeek native web search in milliseconds. | 60000 |
| VERIFIER_PLATFORM_DEEPSEEK_SEARCH_BASE_URL | No | Base URL for DeepSeek Anthropic-compatible search endpoint. | https://api.deepseek.com/anthropic/v1 |
| VERIFIER_PLATFORM_DEEPSEEK_SEARCH_MAX_USES | No | Maximum number of native search uses per request. | 5 |
| VERIFIER_PLATFORM_DEEPSEEK_SEARCH_MAX_TOKENS | No | Maximum output tokens for native search auxiliary round. | 4096 |
| VERIFIER_PLATFORM_WEB_SEARCH_DEFAULT_RESULTS | No | Default number of web search results (5 or 10). | 5 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| logging | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| verifier_platform_verifyA | 把外部主模型生成的回答和完整可审计 reasoning 交给独立 verifier_platform 核查。question、answer、reasoning 均为必填。工具会先检索服务端配置的 UOB 本地知识库；webSearch 支持 enabled、maxResults（5 或 10）和可选 query，联网搜索复用服务端 DeepSeek API Key。返回本地知识库命中、联网来源、confidence、verdict、修正答案及 Ed25519 签名。 |
| verifier_platform_verify_signatureA | 只读、离线验签：使用 RFC 8785 JCS 规范化 payload，验证 verifier_platform 的 Ed25519 签名和可选 SHA-256。不会调用模型。 |
| verifier_platform_statusA | 只读：查看 verifier_platform 的脱敏配置、策略、审计状态和签名公钥。不会调用模型。 |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 3 tools
Each tool has a clearly distinct purpose: full verification with optional web search, offline signature validation, and read-only status/config inspection. The only potential confusion between verify and verify_signature is resolved by the descriptions and by the fact that one calls models while the other validates cryptographic signatures.
All tool names share the verifier_platform_ prefix and use snake_case, making the namespace predictable. Minor inconsistency: status is a noun rather than a verb, so the naming pattern is not consistently verb_noun.
Three tools is a tight, well-scoped set for a verification-focused server. Each tool covers a necessary role: perform verification, verify output integrity, and inspect server status.
The server covers the full lifecycle of a double-check workflow: generate a signed verification result, independently validate that result's signature, and query platform status/config. No obvious missing operation is implied by the stated purpose.