ai-visibility-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AIV_TIMEOUT_MS | No | Timeout per attempt in milliseconds. | 45000 |
| GEMINI_API_KEY | No | Enables Gemini. At least one engine API key must be set. | |
| GOOGLE_API_KEY | No | Alternative to GEMINI_API_KEY; enables Gemini. At least one engine API key must be set. | |
| OPENAI_API_KEY | No | Enables ChatGPT. At least one engine API key must be set. | |
| AIV_CONCURRENCY | No | Maximum number of engine calls in flight at once. | 4 |
| AIV_MAX_RETRIES | No | Number of retries on 429, 5xx, timeouts and network errors. | 2 |
| AIV_GEMINI_MODEL | No | Gemini model to use. | gemini-2.5-flash |
| AIV_OPENAI_MODEL | No | OpenAI model to use. | gpt-5-mini |
| AIV_MAX_QUESTIONS | No | Maximum number of questions per tool call. | 12 |
| ANTHROPIC_API_KEY | No | Enables Claude. At least one engine API key must be set. | |
| PERPLEXITY_API_KEY | No | Enables Perplexity, which returns cited sources. At least one engine API key must be set. | |
| AIV_ANTHROPIC_MODEL | No | Claude model to use. | claude-haiku-5-5 |
| AIV_PERPLEXITY_MODEL | No | Perplexity model to use. | sonar |
| AIV_CACHE_TTL_SECONDS | No | How long to reuse identical question answers, in seconds. | 600 |
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
} |
| prompts | {
"listChanged": true
} |
| resources | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| ask_enginesA | Ask one customer-style question (for example "best HVAC company in Bel Air MD") to every configured AI engine and return each raw answer plus any cited sources. Use this to read exactly what an assistant tells customers. |
| check_visibilityA | Run customer questions across the configured AI engines and report, per engine and per question, whether the business was named, its position in the list, a short evidence quote and which competitors were named instead. Handles name variants (LLC/Inc, "&" vs "and", spacing, small typos). Pass aliases for nicknames or old names. |
| suggest_questionsA | Generate 8 to 12 realistic questions a customer might ask an AI assistant when looking for this kind of business in this city. Uses the first configured engine to write them, or a built-in template set if none is set up. |
| compare_competitorsA | Run customer questions across the configured AI engines and rank which businesses get named most, with how often, by which assistants and at what average position. Pass the business name to flag it in the ranking. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| visibility_report | Walk through a full AI visibility audit for one business and write it up. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| engines | Which AI engines this server can query, the model each uses, and the run limits. Never includes keys. |
TDQS
Scored across 4 tools
Each tool targets a different output: raw answers (ask_engines), per-question business presence (check_visibility), question generation (suggest_questions), and competitor ranking (compare_competitors). The boundary between ask_engines and check_visibility is thin since both fan out the same questions to the same engines, differing only in whether raw output or analysis is returned.
All four tools follow a clean verb_noun snake_case pattern (ask_engines, check_visibility, suggest_questions, compare_competitors). No mixed conventions or vague single-word names.
Four tools is a tight, purposeful surface for a niche AI-visibility monitoring workflow, and each maps to a distinct step. The count is slightly lean given that raw-answer retrieval could arguably be a mode of the analysis tool rather than a peer.
The core loop (generate questions, check presence, benchmark competitors) is covered, but there is no tool for managing configured engines/business identity, no historical or trend tracking across runs, and no export/reporting operation. Agents will hit a dead end if they need to inspect or change configuration rather than just consume results.