Skip to content
MCP Chats

Your MCP server / Personalised by you

Your MCP server, personalised for your work.

Switching agents shouldn’t mean rebuilding your setup or explaining your data from scratch.

Give your supported agents access to the same personalised MCP server: your data and terminology, permitted tools and published processes. Configure it once; connect it where you work.

Join the waitlistTry the permissions
Setup guides for 19 supported clients

Shaped around your work

  • Your records / selected database reads
  • Your actions / selected API endpoints
  • Your terminology / field meanings and notes

Your personalised MCP server

Your context and permitted tools
Your published processes
Your review points and run records

Configure your MCP server once; connect each supported client. Access follows the permissions you grant. Illustrative architecture.

Personalise what the agent can do

A toolkit with the boundaries you choose.

This example lets the agent read projects. Choose whether it can create a task, then see how the process’s access changes the result.

01 / Define the access

Project API

Illustrative imported REST source

GET /projectsRead enabled

Credentials stay in MCP Chats

02 / Try it from the agent’s side

// Request through your MCP endpoint

source: project_api

process: project_process

operation: GET /projects

Awaiting call

Try a tool call. Then change its permissions and try again.

Interactive illustration · These calls use example data. No live systems connected.

Your context, made useful

An agent needs to know what your data means.

Your sources

Connect PostgreSQL, MongoDB, SQL Server or Cosmos DB, or import a REST API. Select the records and actions relevant to your work.

Your boundaries

Give each process the sources it needs. Database tools are read-only; you enable API writes individually.

Your terminology

A status code, an internal name, a field with a specific meaning. Add descriptions, sample values and notes so the context is available when your agent uses the source.

st_cd: “A” = active, “P” = pending

Illustrative access key / Collaborator

Research brief only

Your granted tools. Your configured approvals.
Pause or revoke access when needed.

Share a process without sharing the passwords.

Publish selected processes for a collaborator, client or partner. Their agent gets scoped access; the runs and tool-call records stay with you.

They never receive your source credentials.

Before you build

What makes my MCP server personalised?
You select the data and actions, document their meaning, and configure your processes, required outputs, reviewers and email templates. Your MCP server makes those choices available to connected agents according to their access. Personalisation comes from what you configure; it does not depend on automatically learning your chats.
Where can I use it?
Connect supported coding agents, personal agents, editors and desktop apps through MCP. Each client has its own setup; the each reaches your published processes and granted sources according to its access. See all 19 client setup guides. ChatGPT and claude.ai connectors are not enabled in the current pilot.
Which formats and authentication methods are supported?
REST imports support OpenAPI 3.x, Swagger 2.0 and Postman v2 JSON. Authentication supports none, basic, bearer, API key and OAuth 2.0 client credentials. Setup tests the connection, lets you select resources and document them, then saves a named source. It becomes useful to an agent only after a process grant.
What makes an API write callable?
Import proposes reads for GET and HEAD, and writes for other methods; it never authorises access. An admin must enable an endpoint as a write and grant writing to the process. An admin can classify a POST search endpoint as read-only. Enabling a write is a configuration permission, not a human approval on every call. Agents use granted sources while working on a process step.
What happens to credentials and query results?
Credentials are encrypted at rest and are not sent to agents or logs. PostgreSQL and SQL Server users with write access are refused; MongoDB and Cosmos DB are limited to read operations. Results have row, size and time limits, and partial results are labelled. API origins must be approved; private network addresses need admin permission. Test runs do not call your live systems.
What is recorded, and what if a call’s outcome is unclear?
Reads and writes record the source version, time and fetched or changed data. Steps can require these records as proof. An unknown write outcome stays unknown: it is neither automatically retried nor assumed successful. Published keys can be limited to chosen processes and capabilities, paused or revoked; an organisation-wide token switch is also available.

Current pilot limits. Database writes, YAML imports, other OAuth flows, custom server names and addresses, and proxying existing third-party MCP servers are not available yet.

Connect the rest of the process

ProcessesBuild a routine around your way of working.ApprovalsPause for a person’s review where you choose.Emails in your wordsFill your published wording with each run’s details.

Bring your way of working to your favourite agent.

Join the waitlist with one recurring job in mind. When invited, we’ll help you build the process and try it with your agent. Work solo or with a team; the pilot is free.

Join the waitlist