--- title: "gaja MCP | Teach AI your company’s way of working" description: "gaja MCP connects your definitions, playbooks and data to Claude, ChatGPT and Grok, and lets your AI agents finally know your company." url: https://www.trygaja.com/platform/mcp last_modified: 2026-11-07 --- gaja MCP # Teach your company’s way of working to Claude, ChatGPT, Cursor, Grok, Gemini gaja reads your data and definitions, maps your process and gives it to your AI through MCP, so everyone on the team runs the same playbook every time, in the AI tools they work in today. [Talk with us](https://www.trygaja.com/contact)[Discover gaja Brain](https://www.trygaja.com/platform/brain) The other side: gaja Brain, what your AI reads Works with the AI tools your team uses Claude, OpenAI, Gemini, Copilot, Cursor, Grok, Claude Code, Codex, OpenClaw, Hermes Agent, Perplexity, DeepSeek Why an MCP ## One shared brain, wherever your team works Any AI model connects to gaja through MCP. Your company knowledge is codified and kept in gaja, not in the model or the tools, so you can change any of them and keep it. Claude - Claude - ChatGPT - Gemini - Your own model Context from gaja Thornbury Bakeries Playbook · ICP B #wins Westbrook 4 sources · 2 citations Answer · Claude Open Thornbury Bakeries on cold-chain compliance1, then share the Westbrook Provisions result2: 14% less waste in two quarters. LLM agnostic ### Move to a better model anytime Use Claude for one task and GPT for another, a light model for quick jobs and a stronger one for hard ones. Change the model in gaja, and your playbooks, learnings and skills come with it. Calls to gaja · today All Blocked Access control ### The right access for each person You choose which parts of gaja each person and each agent can read, and which skills they can run. gaja knows who is asking, blocks what they may not see and logs every request. Portable ## Change your tools, keep the memory Connect your stack and gaja learns from every tool in it. Replace any of them, and the deal history, the experiments and what worked are already there for the team and its AI. 01 Change the stack - HubSpot - Gong - Slack - Notion - Apollo 02 Keep the memory gaja · memory kept across tools - Deal history why 214 deals were won or lost HubSpot - Call learnings 38 objections, the answers that worked Gong - What worked 12 plays from the wins channel Slack - Playbooks ICP B, v7 Notion - Targeting the titles and triggers that got replies Apollo 03 Skip the ramp-up - Sales reads gaja in Claude - Customer success reads gaja in ChatGPT - RevOps reads gaja in Gemini Fast ## Each task loads only the context it needs gaja works by progressive disclosure: for the task in front of it, your AI gets a short summary first and opens the full record, with its source, only when it needs more. outbound-agent · mcp Example ### Sharper answers The model reasons over a few conclusions, each with its source, in the part of its context window it handles best. ### Fewer tokens A task loads hundreds of tokens of conclusions instead of the whole account, so every run costs less and starts sooner. ### Room for long work The context fills with what the task needs and stays far from the limit, so long tasks run to the end. ## Give AI the full picture We’re working with a small group of B2B sales teams to connect their tools and put their working methods into gaja. Work email Talk with us Questions ## Questions we get Not here? Ask us directly and we will answer. [Talk with us](https://www.trygaja.com/contact) ### What is an MCP server for sales teams? An MCP server for sales teams is the server a team's AI tools connect to, so Claude, ChatGPT or an agent reads the company's accounts, method and definitions before it works. MCP is the open protocol AI tools use to call outside tools and data. gaja MCP is that server for revenue teams, and each person keeps the AI tool they already use. ### Which AI tools work with gaja MCP? Any AI tool or agent that connects to an MCP server works with gaja MCP: Claude, Claude Code, ChatGPT, Codex, Gemini, Copilot, Cursor, Grok, Perplexity and DeepSeek among them, and a model you run yourself. Each person keeps the tool they prefer, and all of them read the same context, so a rep in ChatGPT and a manager in Claude work from the same accounts and definitions. ### Can we switch from Claude to ChatGPT without losing our company context? Yes. gaja sits outside every AI tool and every model, and each of them reads it through MCP. When a better model ships, the team switches, and the ICP, the playbooks, the skills and every lesson stay where they are. The same holds for a new CRM, call recorder or chat tool: the history and what worked are there from the first day. ### Can an AI agent update our CRM safely through MCP? Yes. The skill writes into the CRM through the CRM's own connector, such as a brief on the deal after a call, an enriched field or a normalised loss reason. Every call an AI tool or agent makes to gaja carries its own key and is logged, by person and by skill, so the brain manager sees what was read and what was changed. ### Does an MCP connection use fewer tokens than pasting documents into the chat? Yes. gaja hands the AI tool conclusions with their sources instead of whole records, so a task loads hundreds of tokens instead of every file and transcript on the account. Each run costs less and starts sooner, and the context window stays far from its limit, so long tasks run to the end. ### Does gaja MCP work for agents that run on their own? Yes. gaja MCP serves an agent the same way it serves a person: the agent has its own identity and key, runs the team's skills on the same accounts and definitions, and every call it makes is logged like anyone else's. Claude Code, Codex and the agents a team builds itself connect the same way. --- What gaja is, and every page an agent can read: https://www.trygaja.com/llms.txt