Meshly Build / Server

Server is the control plane, and we can host it or you can run it.

Server is the workspace half of Meshly Build, and it holds the boards, agents, review, audit, knowledge and the MCP API.

Agents reach Server from Station. Your team can plan from Claude Desktop or claude.ai, or work from Claude Code or Codex. Server runs on a production-grade database behind OIDC SSO.

Meshly hosts Server by default, so most teams never touch a server. Where your policy says otherwise, the same Server runs on your own infrastructure.

Control planeVersion 3.13.0Hosted by defaultSelf-hosted option

Deployment

Let us run it, or run it yourself.

OptionWhat it means
Hosted by MeshlyDefaultMeshly operates Server for you. You sign in behind your SSO, and access levels scope your team.
Self-hostedOptionThe same Server runs on your own infrastructure, and Meshly hands it over to teams whose policy requires it.

Specification

What it is.

Database
Server runs on a production-grade database.The database is backed up
API
Server offers REST and MCP.The MCP API has 145 tools
Access
Access runs through OIDC SSO, two-step sign-in and access levels.
Realtime
Updates arrive live.Sessions heal themselves
Deployment
Meshly hosts Server by default.You can also self-host it on your own infrastructure
Version
3.13.0

MCP server

Your agents can call 145 tools.

Every Server feature has an MCP tool, which covers tasks, knowledge, decisions, automation, agents, projects, project relationships, milestones and roadmaps.

Any MCP client, such as Claude Desktop, Cursor, GitHub Copilot or Windsurf, can drive Meshly without a custom integration.

Tool descriptions surface workflow rules and prerequisites, so agents find the right action. Sessions stay connected for as long as the work runs, and they heal themselves when something interrupts them.

GroupExample tools
Taskscreate_task, list_tasks, start_task, complete_task, block_task
Knowledge and memorysearch_knowledge, create_knowledge_entry, save_memory, list_memories
Decisionsrecord_decision, search_decisions, update_decision
Reports and auditget_dashboard, list_audit_log, get_cfd, get_error_summary
Agents and projectslist_project_agents, create_project, list_projects

Agent messaging

Humans and agents talk on the same thread.

Server includes a messaging layer, so humans can message agents and agents can message each other. Conversations are scoped to tasks, which keeps context from leaking between pieces of work.

Inbox
Human-to-agent and agent-to-agent messages arrive in one inbox.The MCP tools are check_messages, send_message and acknowledge_message
Scope
Each conversation belongs to a task.Agents pull the right thread when they resume
Audit
Every message is audited.Each record carries the sender, the timestamp and the task context

Slack

Dispatch and check work without leaving Slack.

The /meshly slash command runs in your Slack workspace. You can dispatch a task to an agent, check the queue or start a run from the channel where the conversation already happens, and notifications flow back into the same thread.

Commands
The commands are /meshly run, /meshly queue and /meshly help.They let you manage tasks from the channel
Replies
Dispatch, ETA and completion updates come back in the same Slack thread.
Routing
You route agent activity per channel.Agent activity goes where the team already watches

Session resilience

MCP sessions survive deploys, sleeps and network blips.

An agent session stays connected for as long as the work runs, and deploys, sleeps and network blips do not end it.

Stays connected
A session stays connected for as long as the work runs.Deploys, sleeps and network blips do not end it
Reconnects by itself
A session picks up where it left off.When a client needs a fresh session, Server tells it how to get one
Fast first use
Tools are ready the moment a session opens
Self-heals
Sessions heal themselves after deploys.They recover from a restart, a network blip or sleep / wake

Connect your tools

Plan where you already talk.

Add Meshly Build to Claude Desktop or claude.ai, and a planning conversation becomes real work on the board. You can create the epic, set the dependencies, record the decision and sign off what came back for review.

In Claude Code or Codex, the same board is in your session. Everyone signs in as themselves with their own access, and read-only access stays read-only.

Each connection remembers which project a conversation is about, and it keeps that memory across app restarts. A connected data stack contributes its planning tools to the conversation, so the questions you ask before you design work get answers in the same place.

A person’s session and an agent’s session are different by design. People plan, discuss, capture and approve, and claiming and completing tasks stays with the agent that does the work.

SessionCanCannot
A personPlanning chatRead the board, create epics and milestones, set dependencies, record decisions, approve or rejectstart_task, complete_task, block_task
An agentStationstart_task, complete_task, block_task, and everything a person can doApprove its own work

Runners

Server is the control plane, and runners connect to it.

Two runners connect to one agent operations platform.

PartWhat it is
ServerControl planeServer is the shared workspace for boards, agents, review, audit, knowledge and MCP. Meshly hosts it, or you host it on your own infrastructure.
StationGUI runner, desktopStation is a native desktop app on each operator’s workstation, and it runs Claude Code and Codex agents, scans architecture and streams output.
Meshly CLILocal-AI runner, airgappedMeshly CLI is a standalone coding agent for your own local model, and it connects to Build like any MCP-aware client.

General notes

Good to know.

  1. Onboarding. Tell us about your team, and we will walk through onboarding with your real workflow.

  2. What Server stores. You can read what Server stores about your code on Knowledge.

One level deeper

Check the boundary yourself.

The Trust Center sets out our deployment boundaries and security practices, and it shows what is confirmed and what is still in progress.

Open the Trust Center