$ loading_
Use Cohere models for chat, embeddings, reranking, classification, and summarization.
The material is sparse, but the description indicates this is an MCP server for the Cohere AI API, which typically implies local execution and sending user content to an external AI service. Because the source is open and MIT-licensed, the overall posture is mostly caution rather than high risk; the main concerns are missing documentation and unclear endpoint/auth details.
The material says there are no keys/environment variables, yet the tool is described as a server for the Cohere AI API, so the real authentication model is unclear and may rely on DAuth or runtime credential passing. There is no explicit sign of credential abuse, but the auth flow is not transparent and should be checked for token receipt, storage, or forwarding.
Although the metadata says there are no remote endpoints, the functionality explicitly targets the Cohere AI API, which normally means prompts, text, or embedding requests are sent to an external model service. No concrete domains, data-flow details, or configuration scope are provided, so this looks like ordinary network egress with limited transparency.
The system marks it as executes-code, indicating this MCP tool runs locally as a server process and handles requests. That is a normal capability for MCP tools; the material does not show extra privilege escalation, unrelated system command execution, or obvious overreach.
The material does not claim access to specific local files, databases, or system resources, so there is no clear sign of overbroad authorization. However, as a locally running MCP service, it will at least handle user-supplied conversation/text content; if forwarded to Cohere, that data leaves the local machine.
Positive factors are that it is open source, auditable, and MIT-licensed, which materially lowers the risk rating. Points to watch are that it comes from a third-party registry, has 0 GitHub stars, unknown maintenance status, and no README, so supply-chain trust signals are weak but not enough on their own to justify a high-risk rating.
Copy the install command and let the AI configure it · recommended for beginners
No copy-paste install info for "cohere-mcp" yet — see the docs or source repo.
Use Cohere to summarize the following long text, extract 5 key points, and provide a summary within 100 words: {{article content}}A structured summary with five key points and a concise overview.
Rerank the following candidate results by relevance to the user query "{{query}}", and explain why the top 3 items rank highest: {{candidate list}}A reranked results list with brief explanations for the most relevant items.
Use Cohere to classify the following customer feedback into "positive, negative, neutral, feature request, complaint" and output a label and reason for each item: {{feedback data}}Per-item classification results with labels and reasons, ready for further analysis.
Use Perplexity models for web search, chat, embeddings, and content moderation.
Read, edit, and summarize documents through a Claude-powered chat interface.
Build, debug, and manage software tasks with natural language across LLMs.
Delegate low-risk tasks to a cheaper model with main-agent review.
Access multiple AI providers in one terminal for generation, search, and comparison.
Connect DeepSeek language models through MCP for unified app and agent use.