Overview

The Enterspeed Management MCP Server turns the Enterspeed Management API into a set of Model Context Protocol tools that AI clients (Claude, Cursor, VS Code, custom agents) can discover and call. It is the configuration counterpart to the Query MCP server: where Query MCP reads your data, Management MCP reads and changes your setup. You do not need to host anything yourself — Enterspeed runs the server for you. Typical uses:

Ask an agent to explain how a tenant is wired together — sources, environments, domains, clients, deployments.
Have an agent write a mapping schema, dry-run it against a real source entity, and save a new version.
Investigate a view or route that is not producing what you expect.
Query your tenant logs and metrics in natural language.

Hostname

Environment	URL	Transport
Production	`https://mcp.management.enterspeed.com/`	MCP Streamable HTTP

A health check is available without authentication:

curl -s https://mcp.management.enterspeed.com/health
# "ok"

The server speaks MCP Streamable HTTP on the root URL. Responses are delivered as a server-sent event stream on that same URL, so clients that label their transport SSE generally work when pointed at the root. There is no separate /sse endpoint — always configure the root URL.

How authentication works

Authentication in the Management MCP server works through OAuth using your existing Enterspeed account. Once the Management MCP server is installed, a browser window will open and you sign in using your existing Enterspeed login credentials. When you are authenticated simply tell your AI client which tenant you want to work on if you have access to more then one tenant.

What the server can and cannot do

The server exposes 35 tools: 22 reads and 13 additive writes.

Reads cover tenant overview, sources and source entities, mapping schemas, deployments, views and routes, indexes and index documents, domains and hostnames, environment clients, logs, and metrics.
Writes are additive only — create and update. They cover schema authoring (create a schema, save a version, deploy), environments, source groups and sources, domains, and environment clients.

There are no delete tools, by design. This is not a policy the server can be talked out of — it has no way to express a destructive call. Every request the server can make is a GET, or a POST/PUT that appears on a fixed allow-list checked when the server starts. No tool can delete a schema, environment, source, domain, client, or view.

The practical consequence: an agent connected to this server can create clutter you may need to tidy up manually, but it cannot destroy your configuration or your content. See Tools for the full inventory with each tool marked read or write.

Guardrails

Beyond the key’s own permissions, the hosted server applies three limits you should know about.

Schema deploys are restricted to non-production environments

On the hosted server, deploy_mapping_schema only deploys to environments named dev, development, test, or staging. A deploy targeting any other environment — including anything named prod, production, or live — is refused with a deploy_not_permitted error naming the environments that are allowed.This is a restriction of the hosted server, not of your key. Renaming an environment to get around it does not work either: update_environment refuses a rename that would move an environment onto the allowed list.Production deploys stay a deliberate human action in the Management App.

Access keys are never returned through MCP

The server strips the accessKey field out of every response before your client sees it, at every level of nesting. This covers environment client keys and source ingest keys.It applies even to create_environment_client, which mints a key: the tool creates the client, and the key value is not in the response. The agent is told a key was created and that it must be collected from the app — so it reports the right next step instead of inventing a value. Fetch the key in the Management App when you need it.The field is removed rather than masked, so no tool can observe a key at any point.

Rate limit — 60 requests per minute per key

Requests are counted per API key over a rolling one-minute window. Exceeding the limit returns:

HTTP 429 Too Many Requests
Retry-After: <seconds>

Too many requests. Retry after <n> seconds.

Well-behaved MCP clients back off and retry. The limit applies to the whole MCP endpoint, and /health is exempt.For comparison, Query MCP allows 120 requests per minute per key.

Security model

If you are reviewing this server before pointing an agent at your tenant, this is the short version:

No server-held credentials. The server has no Enterspeed key of its own and no way to configure one. It can only act with a key you send it.
Passthrough authorisation. Every permission decision is made by the Management API against your key, not by the MCP server.
No destructive verbs. The server cannot issue a DELETE or a PATCH at all.
Secrets are not readable. Access keys are removed from responses at the HTTP boundary, so no tool can observe one.
Production deploys are out of reach on the hosted server.
Least privilege is yours to set. Use a read-only key unless the agent genuinely needs to write.

Connecting a client

Claude

In Claude, open Settings and then select Connectors either in Claude desktop or on https://claude.ai/new#settings/customize-connectors.Click Add custom connectors and fill out a name (e.g. Enterspeed Management MCP) and the Remote MCP server URL (https://mcp.management.enterspeed.com/).

Once you click Add a browser window opens and you authenticate using your Enterspeed credentials. The MCP server is now connected.

Sample prompts

Use these to smoke-test a fresh connection. Orientation

Give me an overview of my Enterspeed tenant — how many sources, environments, and domains do I have, and how are they connected?

Schema investigation

List my mapping schemas and show me the current version of the one that produces my product pages. What triggers it?

Authoring loop (needs a write-capable key)

Write an index schema that indexes my article entities with title, publish date, and author. Dry-run it against a real article first, then save it as a new version.

Troubleshooting

The route for /products/red-shoe is not returning what I expect. Inspect the route and the view behind it and tell me what is wrong.

Logs

Show me any processing errors in the last 24 hours, grouped by schema.

Troubleshooting

Symptom	Likely cause	Fix
`401 Unauthorized` on every tool call	Wrong key type — an environment client key instead of a Management API key	Use a key starting with `management-`. Environment client keys belong to Query MCP.
Reads work, writes return a permission error	The key is read-only	Expected. Use a write-capable key only if the agent genuinely needs to create or update configuration.
`deploy_not_permitted` when deploying a schema	The target environment is not one the hosted server may deploy to	Deploy to `dev`, `development`, `test`, or `staging`, or deploy to production from the Management App.
`create_environment_client` succeeded but there is no key in the response	Working as intended — access keys are never returned through MCP	Fetch the key in the Management App.
`429` with a `Retry-After` header	More than 60 requests in a minute on this key	Let the client back off, or split work across keys.
Client cannot connect at all	Configured with a `/sse` path, or a proxy that buffers streamed responses	Use the root URL. Streamed responses must pass through unbuffered.

Next steps

Tools — the full tool inventory, read and write.
Query MCP — query your Enterspeed data instead of your configuration.

Schema reference

API

Integrations

Tutorials

Best practices

CLI

MCP server

Overview

Hostname

How authentication works

What the server can and cannot do

Guardrails

Security model

Connecting a client

Sample prompts

Troubleshooting

Next steps

​Hostname

​How authentication works

​What the server can and cannot do

​Guardrails

​Security model

​Connecting a client

​Sample prompts

​Troubleshooting

​Next steps

Hostname

How authentication works

What the server can and cannot do

Guardrails

Security model

Connecting a client

Sample prompts

Troubleshooting

Next steps