A FastAPI-based multi-tool MCP server that integrates web search, document processing, embeddings, and LLM capabilities with support for various providers (Google, OpenAI, Groq, local models). Provides tools for web search, document summarization, Reddit/YouTube search, mapping, git operations, and knowledge base creation.
CoexistAI exhibits significant definition quality gaps across most tools. While tool names follow verb conventions reasonably well (get_*, create_, crawl_), descriptions are generic and lack actionability for LLM reasoning. Most tools have minimal input validation constraints and lack documented output schemas. Error handling guidance is absent. The server exposes internal infrastructure details (file paths, folder structures) as tool parameters, which is a design smell. Security concerns are notable: sensitive operations (file system access, web crawling with knowledge base creation) lack permission gates or audit patterns.
Crawl websites and create knowledge bases from the crawled content.
Create a knowledge base from documents or URLs for semantic search and retrieval.
Break complex research into sub-aspects and explore using relevant tools. Use only when deep research is explicitly requested.
Fetch specific file contents from a git repository after structure analysis.
Inspect repository structure to understand layout and organization.
View local folder structures. Excludes hidden/system/cache files.
Use textual addresses to find locations, routes, or points of interest.
Missing output schema documentation. No tools document what fields they return, making it impossible for LLMs to plan downstream calls or extract required data.
Descriptions lack actionable context. 'Flexible Reddit search across subreddits or topics' does not tell an LLM WHEN to use this vs. get_web_search, what the dependencies are, or what to expect back. Average param description length baseline: 72 chars; most tools here are 30-50 chars.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 40 | - | v1 |
Flexible Reddit search across subreddits or topics.
Mandatory validation tool that confirms all tasks are covered and flags any missing information before final answer.
Extract top-k clickable links from a URL based on a query. Use for news, bookings, and interactive sites.
Convert text to speech using Kokoro ONNX TTS.
Primary tool for exploring web content and local files. Works with both web content and local files/folders. Supports PDFs, CSVs, Excels, plain text, images, and folders. Ideal for search, summarization, Q&A, and discovery.
Deep summarization of documents from web or local sources. Best for long PDFs, structured reports, or dense content.
Search and summarize YouTube videos via keyword or URL.
No parameter validation or constraint documentation. 'max_depth' in get_local_folder_tree lacks min/max bounds. 'top_k' in get_topk_clickable_elements has no range. LLMs will pass absurd values (max_depth=10000, top_k=999999) without constraints.
Security risk: file system paths exposed as parameters (get_local_folder_tree, get_web_search with local_mode). No path sanitization hints in descriptions. Sanitize against... path traversal.' An LLM could be tricked into accessing ../../../etc/passwd or other sensitive directories.
No error recovery guidance. Tools lack descriptions of failure modes, retryability, or next steps. Retry? Try a different repo URL? No guidance.
Multiple tools perform similar operations without clear distinction. 'get_web_search', 'get_deep_research_aspects', and 'get_response_check' all appear to summarize or validate information.
Parameters lack type specificity in descriptions. 'voice' in get_tts_audio has no enum or examples of valid values.
No pagination or result limits documented. Web search tools may return hundreds of results.
Destructive operations (create_knowledge_base, crawl_and_create_kb) lack dry-run or confirmation step. Agents make mistakes.' No mention of confirmations or rollback paths.