MCP Studio

Use MCP Studio to build a new task-specific MCP server, or to edit an existing one. Nexla's pre-built MCP servers are built with MCP Studio too, so you can edit them in Studio. You tell Studio the process you want agents to run and the systems that process touches. Studio probes those systems, builds scoped tools from what it finds, and delivers one governed MCP server your agents can call. MCP Studio is available in Express at express.dev.

You don't write code. You describe what you need in plain English, and you review every proposed tool before Studio builds it. When the server is ready, connect to it from Claude, ChatGPT, Cursor or any other MCP client. See Authentication. Each tool call runs as the person who signed in, or as the key's owner when you use a service key, and uses only the credentials in that person's Nexla account. See On-Behalf-Of Access.

Why Task-Specific Servers

If a server exposes every tool available across a system, the agent has to consider all of them on every request. As the list grows, token use goes up, the agent picks the right tool less often, and the server gets harder to govern. A task-specific server includes only the tools its workflow needs. In Nexla's published benchmarks, a task-specific server sends about 1,200 tokens of tool schema per turn, versus about 39,000 for a vendor's native server.

Each task-specific server contains only the tools its workflow needs. When you connect a server directly, your client loads those tools and the agent calls them by name.

How MCP Studio Builds a Server

Studio reaches each system through Nexla's 1,000+ connectors, runs the Agentic Probe to find the objects, fields and permissions your credentials can reach, and builds named tools from what it finds. Nexla routes each tool call, enforces policies and keeps the audit trail.

Prerequisites

• A Nexla account with access to Express.
• Access to the systems your process uses. The credentials can already be in Nexla, or you can connect them during the build.

Build a Server

1. Go to https://express.dev, sign in, and open MCP Studio. Start a new session -- the breadcrumb reads MCP Studio / New session. You chat with the Studio agent in the left pane. The right-hand pane has the Context, Tools, Build and Live Apps tabs.

2. Describe your purpose. Say what the agent should be able to do, which systems and records it needs, and the date window. Studio uses your purpose to pick the systems and to decide how tools are scoped and named. For example:

   Build an MCP server that lets agents read open pipeline
   in Salesforce and flag at-risk deals. Limit it to
   opportunities closing in the next 90 days.

3. Select your systems and credentials. Studio suggests systems based on your purpose and shows a Select Credentials form. Pick an existing credential for each system, add any other system you want included, then click Submit Context. Studio validates each credential and confirms "Credential validated". If a system has no credential, Studio shows a Connect OAuth step so you can connect the account through Nexla.

   warning

   Don't paste passwords, API keys or service keys into the Studio chat. Credentials belong in Nexla, which keeps them away from agents.

4. Wait while the Agentic Probe checks your systems. When Studio shows a question card, select an option to confirm what to include.

5. Review the proposed tools. Studio lists the tools it verified, the items it left out and why, and anything blocked. To change the set, reply in the chat. For example, you can drop a tool, narrow a filter or change the date range.

6. When the set is right, click Build. Studio then shows progress under BUILD PROGRESS on the Build tab: "Load selected plans and build instructions", "Materialize N tool-ready sources" (N is the number of sources, for example "Materialize three tool-ready sources"), "Creating tools" and "Sync smoke-test and register server".

7. Wait for the Server ready card. It confirms that the server is enabled in this session, shows how many tools it has and which systems it uses, and says the tools are callable in this chat on the next turn.

The Build Tab

The Build tab shows the server's status and connection details, with these sections from top to bottom:

Section	What It Shows
CURRENT SERVER	The server status and the number of deployed tools, for example "Active · 3 deployed tools". Click Refresh to update it.
Server ready	Confirms the server is enabled in this session and shows its tool count and systems.
Integration config	Connection details. Quick Connect has a Google OAuth URL and an Azure OAuth URL, each with a Copy button. MCP Config has an mcp.json block you can paste into a client.
BUILD PROGRESS	The four build steps and whether each one finished.
Deployed tools	Every tool on the server. Click Inspect to see a tool's details.

For step-by-step client setup with these URLs and the mcp.json block, see Authentication.

What Studio Creates

A finished build includes:

• A working MCP server -- Scoped to the systems and purpose you described.
• Governed data -- A schema, documentation and access controls for every source.
• Named MCP tools -- Tools agents can call, each with an LLM-written description.
• Managed auth -- Credentials stay in Nexla; agents never see them.
• Full audit trail -- Every tool call is logged with who, what and when, and gets a receipt. See Audit & Receipts.

The server gets three connection URLs: Google OAuth, Azure OAuth and service key. See Authentication.

Tool names are snake_case and usually start with a verb, such as list_calls or search_company_news. Each tool description is written for agents, so they know when to call the tool and what to pass.

Test the Server in Studio

1. Check that your server is included in the count on the MCPs button in the chat box. Studio enables a new server in your session when the build finishes.
2. In your next message, ask a real question the server should answer.
3. Open the "N tools called" group in the reply to see each tool call and its response.
4. Compare the results with the source system. For example, check that the number of opportunities matches what Salesforce shows for the same filter.

Edit a Server

To edit a server, whether you built it or it's a pre-built server, open it in MCP Studio and describe the change in the chat. For example, add a tool, tighten a parameter or rebind a credential. Each change creates a new revision. The server keeps its URLs, so you don't need to change client configuration. Refresh or reconnect a client to see added or removed tools.

Prompting Tips

• Build one server per business process.
• Name the systems, objects, fields and date ranges you need.
• Ask Studio to report row counts, then compare them with the source system.
• If Studio offers several possible causes for a problem, ask it to test each one and report what it ran.

Troubleshooting

Tools Are Not Callable Yet

New tools become callable on the turn after the build. Send another message, and check that your server is included in the count on the MCPs button. On the Build tab, click Refresh and confirm that CURRENT SERVER shows the server as active with the expected number of tools. If an external client connects but lists no tools, see Authentication.

A Credential Doesn't Work

Pick the credential again in the Select Credentials form, or use the Connect OAuth step to reconnect the account. Confirm that the account can reach the objects your purpose names.

Studio Reports a Source as Blocked

Studio explains why it can't build a tool from a source. Fix the cause and ask Studio to probe the source again, or tell it to leave the source out and continue with the remaining tools.

Next Steps

• Authentication -- Connect your server from Claude, ChatGPT, Cursor and other MCP clients with OAuth or a service key.
• Pre-built MCP Servers -- Use a ready-made task-specific server.
• Audit & Receipts -- Look up the receipt for any tool call.
• Connectors -- Browse the 1,000+ connectors that servers are built from.