advisor-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| ADVISOR_MODEL | Yes | Explicit advisor model ID. Required; the server fails with 'configuration error: Missing ADVISOR_MODEL' if unset. | |
| ADVISOR_STORE | No | Session storage backend: 'memory' or 'file'. Use 'file' when the host may restart the MCP process between consults. | memory |
| ADVISOR_STYLE | No | Default advice style: 'brief' or 'detailed'. | brief |
| OPENAI_API_KEY | No | Accepted as a fallback for ADVISOR_API_KEY in openai-compatible mode. | |
| ADVISOR_API_KEY | No | API key for the advisor provider. Takes precedence over ANTHROPIC_API_KEY and is the primary substitute for OPENAI_API_KEY. Optional for local endpoints; if omitted no Authorization header is sent. | |
| ADVISOR_BASE_URL | No | Base URL of the OpenAI-compatible endpoint (OpenRouter, vLLM, Ollama, LM Studio, Magpie gateway, etc.). Default is https://api.openai.com/v1. | https://api.openai.com/v1 |
| ADVISOR_DATA_DIR | No | Session directory when file storage is enabled. | ~/.advisor-mcp |
| ADVISOR_PROVIDER | No | Provider backend: 'openai-compatible' (default) or 'anthropic'. | |
| ADVISOR_HTTP_HOST | No | HTTP bind address. Use 0.0.0.0 inside a container. | 127.0.0.1 |
| ADVISOR_HTTP_PATH | No | MCP endpoint path. | /mcp |
| ADVISOR_HTTP_PORT | No | HTTP listen port. | 8787 |
| ADVISOR_TRANSPORT | No | Transport mode: 'stdio' (default) or 'http' for remote Streamable HTTP. | stdio |
| ANTHROPIC_API_KEY | No | API key used in anthropic mode; ADVISOR_API_KEY takes precedence over it. | |
| ADVISOR_HTTP_TOKEN | No | Bearer token for HTTP mode; at least 16 characters, preferably 32 random bytes in hex. The server refuses to start without it unless the host is loopback and ADVISOR_HTTP_ALLOW_UNAUTHENTICATED_LOOPBACK=true. | |
| ADVISOR_TIMEOUT_MS | No | Provider timeout per attempt, in milliseconds. | 60000 |
| ADVISOR_TOKEN_PARAM | No | Set to 'max_tokens' for providers that reject max_completion_tokens (fixes HTTP 400 about a token field). | |
| ADVISOR_HTTP_TRUST_PROXY | No | Trust X-Forwarded-For. Enable only behind a trusted reverse proxy. | false |
| ADVISOR_MAX_CONTEXT_CHARS | No | Evidence budget before truncation. | 120000 |
| ADVISOR_MAX_OUTPUT_TOKENS | No | Advisor output cap per call. | 1200 |
| ADVISOR_HTTP_ALLOWED_HOSTS | No | Comma-separated Host hostnames; set this in production. | |
| ADVISOR_HTTP_MAX_BODY_BYTES | No | Maximum MCP request body size in bytes. | 1048576 |
| ADVISOR_HTTP_ALLOWED_ORIGINS | No | Comma-separated origin hostnames, or non-web wildcards such as 'moz-extension://*'. | |
| ADVISOR_MAX_CALLS_PER_MINUTE | No | Advisor calls allowed per session per minute. | 30 |
| ADVISOR_MAX_CALLS_PER_SESSION | No | Advisor calls allowed per session; exceeding it yields limit_exceeded. | 12 |
| ADVISOR_HTTP_REQUESTS_PER_MINUTE | No | Requests allowed per client IP per minute; exceeding it returns HTTP 429. | 120 |
| ADVISOR_HTTP_MAX_CONCURRENT_REQUESTS | No | Concurrent MCP requests allowed per process. | 8 |
| ADVISOR_HTTP_ALLOW_UNAUTHENTICATED_LOOPBACK | No | Permit tokenless HTTP only on a loopback bind (development only). | false |
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
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| consult_advisorA | Get an independent second opinion before committing to a non-obvious plan, after repeated failures, or before declaring complex work complete. Supply a compact evidence pack; the advisor cannot inspect your workspace. Each call counts toward the per-session call budget. |
| get_advisor_sessionB | Return the accumulated goal, constraints, decisions, open questions, attempted call count, and recent consults for a session_id. |
| update_advisor_sessionA | Record durable goal, constraints, touched files, or open questions without spending an advisor call. |
| reset_advisor_sessionA | Delete the stored state and call counter for a session_id. Use when the task is complete or the context has changed fundamentally. |
| advisor_healthA | Report non-secret configuration and verify that the configured provider has credentials. Does not make a model request. |
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 5 tools
Each tool has a clearly distinct role: read state (get), write state (update), clear state (reset), invoke the model (consult), and check configuration (health). The get-vs-consult distinction is well clarified by descriptions, so an agent can reliably pick the right tool.
Four tools follow a verb_noun pattern (get/update/reset_advisor_session, consult_advisor), but advisor_health breaks it with a noun-only, verbless form. The deviation is minor and still readable.
Five tools is well-scoped for a session-backed advisor: one core consult plus a minimal state lifecycle and a health check. Every tool earns its place with no redundancy.
The session lifecycle is covered (get, update, reset, consult, health), with creation implied on first write. Only minor gaps exist, such as no list/enumerate operation for existing sessions.