Canonical: https://molo17.com/ai-hub/enterprise-brain/connect-your-ai-tools/ Markdown mirror: https://molo17.com/ai-hub/enterprise-brain/connect-your-ai-tools/index.md Title: Connect your AI tools · Gluesync Enterprise Brain · MOLO17 Description: Connect Claude, ChatGPT, Grok, Meta Muse, Hermes Agent, OpenClaw, or Cursor to the Gluesync Enterprise Brain through the Core Hub MCP server and answer company questions under the caller [Gluesync Enterprise Brain](/ai-hub/enterprise-brain/) · Connect # Every AI tool your company uses, one shared brain. Claude, ChatGPT, Grok, Meta Muse, Hermes Agent, OpenClaw, Cursor: they all speak the Model Context Protocol, and Gluesync Core Hub is an MCP server. Point them at your Core Hub and they get immediate answers based on company shared knowledge, with the permissions of the person asking. [Book a demo](/book-a-demo/) [Read the docs](https://docs.molo17.com/gluesync/latest/gs-modules/corehub-mcp.html) 1. 01 Copy the Core Hub MCP endpoint 2. 02 Paste a token from your account 3. 03 Ask the brain from the client you already use How it works ## No proxy, no export, no new identity 1. 01 Endpoint ### Core Hub is already the server Core Hub mounts Streamable HTTP at /mcp/http and SSE at /mcp beside its REST API. There is no sidecar to deploy and nothing to sync to a vendor cloud. 2. 02 Identity ### The caller's token, the caller's role Each client authenticates with a session token or a Personal API Token. Every tool call runs under that user's Core Hub RBAC, so an AI-shaped path around your permissions does not exist. 3. 03 Answer ### Published agents do the work Clients call list\_published\_agents, start\_agent\_run, and get\_agent\_run to hand a question to an immutable Spark agent version, or search\_enterprise\_brain to look up the tables and procedures directly. Capabilities ## Five MCP tools, one door for every client The AI agents category of the Core Hub MCP server is how external tools reach the brain. Runs execute as the caller, results carry the skills they produced, and nothing returns row data unless the caller could already read it. Ask ### search\_enterprise\_brain Search the schema graph and the memories visible to the caller. Returns tables, columns, and procedures: the route to an answer, not the answer's rows. - Keyword search over table and column names plus a foreign-key walk - Embedding search blended in when you configured an embedding model - Private memories of other users are never returned Delegate ### start\_agent\_run · get\_agent\_run Hand a question to a published Spark agent and read back a durable, redacted result with the skills it used or created. - Immutable agent versions, idempotency keys, cancellation - Runs started by an API key execute as the key owner - Writes wait for a signed-in approval; reads flow immediately Contribute ### publish\_skill · list\_published\_agents Let a client discover which agents exist and, with the right scope, publish a procedure as a reusable skill the whole company can call. - publish\_skill needs Manager, Super admin, or the skills:publish key scope - Skills are versioned and immutable; chat links straight to them - Works from Claude Code, Cursor, or any agent framework FAQ ## Connect your AI tools: questions we hear Which clients are supported? Any client that implements the Model Context Protocol over Streamable HTTP or SSE. Claude Desktop and Claude Code, ChatGPT connectors and the OpenAI Agents SDK, Grok bots, Meta Muse, Hermes Agent, OpenClaw, Cursor, Windsurf, and MCP Inspector all qualify. The in-product Spark agents use the same brain natively. Does a Grok bot or Meta Muse get more access than the user who configured it? No. The client sends a Core Hub token, and every call runs with that token's role and Query Studio data-access rules. A bot configured by a viewer sees what a viewer sees. Can an external client change my pipelines or data? Only through tools its token is allowed to use, and only after a signed-in approval for writes proposed by a published agent. Read-only deployments can hide every write tool with a single setting. What does the model provider see? Only what the agent puts in its prompt: the question, up to eight memory lines, up to eight schema-graph lines, and tool results the caller could read anyway. Provider credentials live in your own vault, and you can route to a local model. Enterprise brain ## Keep exploring [Overview **Gluesync Enterprise Brain** The foundation for every AI tool your company uses, built on the databases Gluesync already integrates. Explore →](/ai-hub/enterprise-brain/) [Extend **Skills and company tools** Agents publish the procedures they discover as skills, and approved MCP servers such as GitLab or Notion become tool bundles you govern. Explore →](/ai-hub/enterprise-brain/skills-and-company-tools/) [Own **Ownership and privacy** A dedicated brain database on your infrastructure, memory shared by default and private on request, never rows, never secrets, never SaaS. Explore →](/ai-hub/enterprise-brain/ownership-and-privacy/) Gluesync Enterprise Brain ## Connect the client your team already uses. One endpoint, one token, one brain. Bring Claude, ChatGPT, Grok, Meta Muse, or your own agent framework, and keep every answer under your Core Hub permissions. [Book a demo](/book-a-demo/) [Get Gluesync](/get-gluesync/)