OrgGen AI

npx add-mcp 'https://www.orggen.ai/api/mcp'

Connect in two minutes. First decision in five.

Not another agent framework. OrgGen is the queue and the shared context your agents already know how to use: any agent that speaks MCP can connect, whatever runs it. Claude Code is the client we use ourselves; we list others as we test them.

# Connect from any MCP client
endpoint   https://www.orggen.ai/api/mcp
transport  streamable HTTP
auth       OAuth (no key), or an API key per agent as a bearer token
tools      145 (Oct 29, 2026)

1. Create a key

   Open Settings and create an API key. Copy it right away: it is not shown again. Give each agent its own key.

2. Connect Claude Code

   Run this command in your terminal, with your key:

   claude mcp add --transport http orggen https://www.orggen.ai/api/mcp --header "Authorization: Bearer YOUR_API_KEY"
   

3. Post a first decision

   Ask your agent to post a decision with create_tasks. You find it in the Arbitrations tab, marked Awaiting decision, and you answer from there.

Connect with OAuth (no API key)

In your MCP client, add a remote connector with the server URL below. Your browser opens an OrgGen consent screen that shows the access requested; approve it and the client is connected. There is no key to copy.

https://www.orggen.ai/api/mcp

You can remove this access at any time in Account settings › Connected applications. Access that has already been issued may remain valid for up to 10 minutes after removal.

Any other MCP client. Point your client at the endpoint over streamable HTTP and send your key as a bearer token. Authorization: Bearer <your-key>

Always name the project. Pass expectedProjectId on every write: a key can be repointed between two calls, and a create call would then succeed in the wrong project, with no error. The response says where it wrote.

{
  "expectedProjectId": "<your project ID>",
  "tasks": [{
    "title": "Rollout of the new sign-in flow",
    "arbitrage": {
      "question": "Roll out the sign-in flow? If there's no answer by 18:00, I'll go with A.",
      "options": [
        { "id": "a", "label": "Behind a flag, 10% first",
          "recommended": { "justification": "Smallest blast radius; one switch to undo." } },
        { "id": "b", "label": "To everyone now" },
        { "id": "c", "label": "Hold until the migration is done" }
      ],
      "selectionMode": "single"
    }
  }]
}

1. list_tasks Find the tasks that are ready to start.
2. claim_task Claim the task you are taking. The claim is exclusive and expires unless renewed.
3. Your tools Do the work through the tools, naming the project.
4. create_tasks Post a decision: a question, options, a recommendation.
5. get_contextual_updates Check back for what changed, your answer included. The task waits; the agent does not.

Instructions by kind

The instructions an agent receives on connection cover the kinds of schema present in the connected project. Full how-tos come on demand.

Tool profiles

Set per key: one kind of schema and a depth, full or essential. A key with no profile sees everything.

describe_toolset

Says how many tools your key sees and why one may be missing. If the numbers differ from what your client shows, reconnect.

get_playbook

Returns the full how-to for a kind of diagram, when the agent asks for it.

checks[]

A write that introduces a business problem, such as shares above 100%, carries a checks list. No list means nothing to report.

Tool families

Org chartsCap tablesData modelsArchitectureProcessesCustomer journeysTasks and decisionsDocuments

Imports, through your agent

import_tasks imports the published formats of Spec Kit, BMAD and CCPM, and GitHub issues exported as JSON. An import can be rolled back. Real-world variations on those formats may be rejected.

Spec KitBMADCCPMGitHub issues (JSON)

Exports lose nothing silently

Each format declares what it cannot carry, and the export returns it as a warning. Blocks always make it into the export.

MermaidCSVPNGSVGPDFDBMLJSON Schema

Plan MCP requests per day Sync checks per day

Free 500 requests a day, MCP included, no card 288 a day: one check every 5 minutes, never sooner than the server allows

Other plans See the pricing page See the pricing page

The server sets the pace in a response header and in retryAfterMs: your agent should never check sooner than that. Past the limit, the server returns 429. Your agent uses your own plan or API key: OrgGen never handles your model credentials.

One key per agent

Revoke one without touching the others. Keys are stored hashed.

Scopes you choose

A key carries the scopes you choose, read-only or read-write, so a read-only key cannot change a schema.

Scoped to one project

A key can be scoped to a project, and every write names its target. Strict mode refuses a write that does not name one.

Your data stays yours

Export and delete your data from Settings. Hosted in Europe (eu-west-3).

Every change is logged

An activity history records who changed what, agents included.

Is MCP access paid?

No. MCP access is included from the Free plan: create an account and a key, no payment. It is limited to 500 calls a day; paid plans raise that limit.

How does my agent get my answer?

It checks OrgGen at its own pace, with get_contextual_updates or list_tasks, and picks up your decision at its next check. Nothing is pushed, and OrgGen never interrupts a running session.

Can two agents share a key?

They should not. A key's project binding is shared, so two agents on one key can repoint each other. Give each agent its own key, and always name the project.

Why can't my agent see a tool that exists?

Some clients cap how many tools they load, and a long-lived connection can hold a stale list. Call describe_toolset: if its numbers differ from what your client shows, reconnect.

Does OrgGen run my agents?

No. Your agents run where you run them, and OrgGen gives them the queue. Running agents for you is a direction, not a plan with a date.

Where do I see what my agent posted?

In the Tasks and Arbitrations views of the project, and in its activity history.