Connect Cohere to ChatGPT via MCP: Chat and Search

💡 No install? Use cloud.anythingmcp.com directly. Sign in, click Connectors → Cohere, paste your credentials, mint an MCP API key and you're done. No Docker, no git clone, no local server.

Cohere + ChatGPT

Cohere with your own API key: answers from Command models with citations from your documents, embeddings for semantic search, reranking of search results, speech-to-text, parsing of document images, token counting, and the models, datasets, batches and embed jobs of the account.

Prerequisites

See the full setup instructions baked into the connector (visible in the in-app store when you select the connector). The required environment variables for this connector are:

COHERE_API_KEY

Step 1 — Get credentials

Setting up takes about 2 minutes. You need a Cohere account; a trial key is free.

1. Sign up at https://dashboard.cohere.com and open API Keys (https://dashboard.cohere.com/api-keys).
2. Copy the trial key, or create a production key after adding a payment method under Billing.
3. Paste it into COHERE_API_KEY, without the word Bearer. The connector sends it as Authorization: Bearer and only to Cohere.
4. Install the connector and run cohere_check_api_key or cohere_list_models to check the key.

Which tools cost money: cohere_chat, cohere_embed, cohere_rerank, cohere_transcribe_audio and cohere_parse_document run a model. With a production key every call is billed: chat per input and output token, embed per token or image, rerank per search of up to 100 documents (longer documents count as several). A trial key is free but rate limited per minute and per month and may not be used commercially. The AI asks you before large or repeated calls and sets max_tokens to cap a chat reply. cohere_tokenize, cohere_detokenize, cohere_check_api_key and all list and get tools cost nothing.

Models: ids change, so cohere_list_models (filter endpoint chat, embed or rerank) has the current list and cohere_get_model shows one model's context length, features and endpoints.

• Chat: command-a-03-2025 (the default choice), command-a-reasoning-08-2025 (reasoning), command-a-vision-07-2025 (images), command-a-translate-08-2025 (translation), command-r7b-12-2024 (small and cheap).
• Embed: embed-v4.0 (text and images), embed-v5.0-pro, embed-v5.0-fast, embed-multilingual-v3.0.
• Rerank: rerank-v5.0-fast (default), rerank-v5.0-pro (more accurate), rerank-v3.5.
• Transcription: cohere-transcribe-03-2026. Parsing: parse-v5.0.

Answers with citations: cohere_chat takes model and messages and returns the whole reply at once. Pass documents (plain strings, or objects with id and data holding title and snippet) for a grounded answer: the reply then carries message.citations pointing into them. response_format forces JSON, tools adds function calling, and thinking with type disabled turns reasoning off on reasoning models.

Semantic search: cohere_embed turns up to 96 texts (or images as data URIs) into vectors; input_type search_document is for what you store, search_query for the query. The default here is embed-v4.0 with 256 dimensions (output_dimension 256, 512, 1024 or 1536). cohere_rerank orders up to 1,000 documents by relevance to a query and returns results with index (the position in your list) and relevance_score from 0 to 1; top_n keeps only the best.

Audio and documents: cohere_transcribe_audio takes a public link to an audio file (flac, mp3, mpeg, mpga, ogg, wav, up to 10 MB through this connector) and the required language (ISO-639-1); the connector downloads the file and uploads it to Cohere. cohere_parse_document turns an image of a page (scan, photo, screenshot; public link or data URI up to 20 MB) into markdown with tables inlined, or into layout blocks. PDFs are not supported yet.

Tokens: cohere_tokenize counts the tokens of a text for a model, for example before a chat call; cohere_detokenize turns token ids back into text.

Account data: cohere_list_datasets, cohere_get_dataset and cohere_get_dataset_usage (storage used, at most 10 GB per organization); cohere_list_batches and cohere_get_batch for Batch API jobs; cohere_list_embed_jobs and cohere_get_embed_job for embed jobs.

Data formats: chat returns message.content[] (items of type text with text), finish_reason (COMPLETE, MAX_TOKENS, STOP_SEQUENCE, TOOL_CALL, ERROR) and usage.billed_units with the billed input and output tokens. Embeddings are under embeddings.float, one vector per text in order. Dates are ISO 8601, UTC.

Paging: cohere_list_models and cohere_list_batches return next_page_token; the AI passes it as page_token until it is empty. cohere_list_datasets pages with limit and offset.

Switched off at install: nothing. All 17 tools are on, and none of them deletes anything.

Limits: at most 96 texts per embed call and 1,000 documents per rerank; audio up to 10 MB and document images up to 20 MB. Trial keys have low per-minute and monthly limits.

Common errors: 400 means a bad request (for example a parameter the model does not take, or more than 96 texts); 401 a wrong key; 402 a payment method is needed (add one under Billing); 404 an unknown model, dataset, batch or job; 422 a request that could not be processed; 429 the rate limit (the AI waits instead of retrying in a loop); 499 a cancelled request; 500 and 503 are Cohere problems, retry once later.

The connector does not stream replies and does not create, cancel or delete datasets, batches or embed jobs. Classify, Summarize, Generate, connectors and fine-tuning are deprecated by Cohere and left out.

Step 2 — Install the adapter

curl -fsSL https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml -o docker-compose.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d

Step 3 — Add the connector in ChatGPT

Your server URL: in AnythingMCP, open MCP Servers → the server this connector is on, and copy its URL (https://cloud.anythingmcp.com/mcp/…). Use it wherever this guide shows YOUR_SERVER_ID.

ChatGPT adds MCP servers as plugins on its Plugins page, on the web and on paid plans. Developer mode is no longer needed. In a Business workspace only admins and owners can add one; in Enterprise and Edu an admin can grant the Create plugins with MCPs permission.

1. Go to chatgpt.com/plugins, click Add, then Add custom MCP server.
2. Enter a name, e.g. Name: Cohere.
3. Under Connection, enter the Server URL https://cloud.anythingmcp.com/mcp/YOUR_SERVER_ID and keep Authentication on OAuth.
4. Select I understand and want to continue, then Create as a plugin. When ChatGPT asks, sign in to AnythingMCP and allow access.
5. Install the new plugin from your personal plugins, then type @ in a chat and pick it.

Available tools

Example prompts

• "Answer the question about our refund window using only these three policy excerpts, and show which excerpt each sentence comes from."
• "Which of these 20 product descriptions best match waterproof winter hiking boots? Rerank them and show the top 5 with scores."
• "How many tokens does this contract text have for command-a-03-2025?"
• "Turn the scanned invoice at https://example.com/scans/invoice-0425.png into markdown, including the line-item table."
• "Which of my Cohere embed jobs and batches are still running, and did any of them fail?"

FAQ

Does the free ChatGPT plan work with custom MCP servers? OpenAI documents custom MCP servers for paid plans only; the Free plan is not on its list.

Next steps

• Connector landing page
• Connect to Claude
• Connect to Cursor
• Connect to Gemini
• Connect to Meta Muse
• Connect to Microsoft Copilot