Skip to main content
Connic
Getting Started

Coding Agent Setup

ChatGPT, Claude, and coding agents can access the Connic playbook and permission-scoped platform tools through the full plugin or a separate skill and MCP connection, depending on the client.

Last updated
On this page
Quickstart

Start a new session and ask your AI agent to install the Connic plugin.

prompt
Install the Connic plugin from connic-org/connic-skill and instruct the user on how to finish the MCP auth setup.

Manual Setup

Choose a client for its current plugin, skill, and MCP instructions.

ChatGPT app setup

  1. Add the Connic marketplace

    Open the ChatGPT desktop app, select Plugins → Add → Add a marketplace, then enter connic-org/connic-skill in the Source field.

  2. Confirm the marketplace

    In the Add plugin marketplace dialog, leave Git ref and Sparse paths blank, then select Add marketplace.

  3. Install and connect Connic

    Open the new Connic marketplace, select the Connic plugin, and choose Install. Complete the connection prompt and choose the project, environments, and permissions in Connic.

  4. Use Connic

    Start a new Chat or Work conversation, or a Codex task. Type @ in the composer and choose Connic, then ask it to work with the Connic project.

    See the official OpenAI plugin instructions.

Now tell it what you want to build

Start a new task or conversation with Connic enabled. Describe what your agent should do, or try this example in a coding client.

prompt
Build a customer-support agent in this repository with Connic.
Add the agent configuration, any required tools, and a focused test suite.
Validate the project with connic lint when you are done.

That’s it. You’re ready to build with Connic.

Plugin and skill

The Connic skill provides the authoring playbook: project structure, SDK and CLI workflows, connector patterns, tests, and platform conventions. It does not grant live access. Full plugins bundle the skill with the Connic MCP connection.

The project copies installed by connic skill are skill-only; accepting its optional client prompt installs the full plugin separately. npx skills add also does not change MCP configuration. MCP-only installs add live tools but not the Connic authoring playbook. View the plugin and skill source.

Connic MCP

Connic MCP connects a coding client to https://mcp.connic.co/mcp for live, permission-scoped project state and actions. It does not add tools to a deployed agent or expose an agent as an MCP tool. Review its official MCP Registry listing.

Access and permissions

OAuth narrows access; it never expands the current Connic membership permissions.

Access boundaryHow it works
Project and environmentsEach authorization belongs to one user, one OAuth client, and one project. Choose all project environments or an explicit environment subset during consent.
Read and write permissionsConnic can only grant permissions allowed by the current project membership. Keep all eligible reads or select individual read permissions. Write actions are optional and can be granted together or individually.
Review and revokeManage the connection under Project Settings → API Keys & MCP Auth. Changes to the environment scope and permissions take effect without reconnecting. MCP-triggered project actions follow the same Audit Log behavior as equivalent API actions, while request-level activity stays in service logs. Revocation immediately ends access.

Projects that require two-factor authentication also require a recent two-factor sign-in before sensitive writes.

Available actions

The client discovers only the tools allowed by the current authorization. Stored secret values are never returned.

AreaAvailable capabilities
Scope, stats & environmentsContext, health, aggregate stats, run history, environment lifecycle, defaults, and queued runs.
Agents & runsAgent details, runs, traces, inputs and outputs, logs, cancellation, reruns, and bulk operations.
DeploymentsDeployment details, tests, runs, creation, cancellation, activation, and redeployment.
ConnectorsConnector definitions, agent links, runs, stats, and logs.
RetrievalNamespaces, entries, chunks, statistics, uploads, deletion, sources, and sync runs.
DatabaseEnvironment-scoped stats, collections, inferred schemas, full document queries and counts, updates to existing rows, and document and collection deletion.
SessionsList, search, and filter environment-scoped persistent sessions with stored state, agent and user identity, timestamps, and pagination; delete individual sessions or matching sessions in bulk.
Approvals & channelsPending decisions, approval configuration and routing, and notification channel management.
Budgets, audit & BridgeCost summaries, rankings and alerts; audit views; Bridge routes and route updates.
Judges & A/B testsConfigurations, results, statistics, variants, plans, comparisons, lifecycle, and evaluation queues.
AI GovernanceSystems, assessments, controls, transparency, monitoring, incidents, and evidence snapshots.

Connic MCP does not start agent runs directly and does not expose payments, credential management, project deletion, or account-security operations. Use connectors for event-driven runs. Treat text returned from runs, logs, connectors, sessions, and retrieval as data, not instructions.

Continue with Connic

REST API

Use project API endpoints directly from custom software

CLI

Build, test, and deploy from the terminal

Runs & Traces

Understand the run data exposed by MCP tools

Approvals

Review and control tool calls that require approval