# Connect your Paperclip agents to Tareno

> Source: https://tareno.co/agents/paperclip
> Language: en
> Paperclip schedules agents but adds MCP servers in the runtime it launches. See where to configure Tareno for Claude Code or Hermes agents, how OAuth works with heartbeats and what to verify.

Paperclip is a control plane: it organises agents into a company with goals, budgets and scheduled heartbeats, and launches a runtime for each. Its documentation is direct about MCP: Paperclip has no command for adding a server, because the control plane launches the runtime and the runtime is what speaks MCP. For Tareno that means the connection is configured in Claude Code or Hermes, whichever adapter an agent uses, and then every heartbeat inherits it. This page explains where to configure it, how sign-in survives unattended runs and what is not yet documented. Tareno has not tested it end to end.

## Connection facts

| Field | Value |
| --- | --- |
| Product surface | Paperclip (paperclip) |
| Available connection | Custom MCP connection |
| Connection evidence | documented |
| Methods | mcp |
| Authentication | client_dependent |
| Setup | [Setup guide](https://tareno.co/docs/mcp) |
| Install or connect | [Open setup](https://tareno.co/docs/mcp) |
| Evidence observed | 2026-10-25 |
| Facts reviewed | 2026-10-25 |

Publication status and a working connection are separate facts. Evidence for one product surface does not establish installation or compatibility in another surface.

## How do I connect Tareno to a Paperclip agent?

Configure it in the agent's runtime. For a Claude Code adapter, add Tareno as an http server in user scope and sign in once as the same operating-system user that runs the heartbeat. For a Hermes adapter, enable the mcp toolset and define the server in Hermes's own config. Paperclip's documentation covers only these two adapters, and Tareno has not tested either with it.

## Connect Tareno in four steps

1. **Pick the adapter that has documentation.** Paperclip's guide documents MCP for the Claude Code and Hermes adapters. It expects other CLI adapters, such as Codex, Cursor Local or Gemini CLI, to inherit MCP from their own tools, but it does not cover that path yet, so do not assume it for them.
2. **Add Tareno in the runtime.** For Claude Code, the guide uses the claude mcp add command with the http transport and user scope for remote servers, because OAuth tokens should not sit in git. Use the address shown on the right. For Hermes, set the transport to http and a url under the server entry, and include mcp in the adapter's toolsets or the servers are ignored.
3. **Sign in as the heartbeat's user.** The guide says to complete the browser flow once as the same operating-system user that runs the heartbeat, using the /mcp command in an interactive claude session. Tokens persist and refresh. For Hermes, Tareno documents the hermes mcp login tareno command.
4. **Verify from the agent's directory.** Run claude mcp get tareno in the agent's working directory with the environment the heartbeat uses, and look for OAuth configured. Then have the agent call `list_accounts` in a run. The run viewer shows which tools were available, which is the check that the connection reached the agent.

Check the server from the agent's directory:

```bash
claude mcp get tareno
```

Tareno's remote MCP endpoint:

```json
"url": "https://tareno.co/api/mcp"
```

## Where Paperclip helps with Tareno

Paperclip is about many agents working toward goals on a schedule. These jobs use that structure, and each has a workflow guide.

### An agent workstream becomes a progress post

When an agent has been working a goal for a week, the record exists in Paperclip: issues closed, budget used, what is blocked. A reporting agent can turn that into an X post about the work and save it as a draft, which you read against the tickets before any time is requested.

### A manager's retrospective on a LinkedIn draft

A company of agents produces a lot of activity that a human manager wants to learn from. A retrospective agent can summarise what worked and what failed over a month into a LinkedIn draft in the manager's voice, with the human deciding what to disclose publicly.

### Budgets and approvals around the whole thing

Paperclip tracks cost with monthly budgets and Tareno holds publishing as a pending action. A reporting agent that drafts weekly stays inside both limits, because its runs are small and its writes are drafts, and a person remains the only one who can turn any draft into a published post.

## Prompts to copy

### Check the connection

```text
Use Tareno to list my connected social accounts with platform and account name. Read only: do not create, schedule or publish anything. Tell me which account you would use for a progress post on X and which for a monthly LinkedIn retrospective.
```

### Progress post from tickets

```text
Read the closed issues of this goal from the last seven days and write an X post within 280 characters about the result that matters most to an outsider. List the issue numbers the claim rests on, and do not mention anything that is still blocked.
```

### Monthly retrospective

```text
From this month's run records and budget summary, write a LinkedIn draft of about 220 words for the manager: what worked, what failed and one change for next month. Remove internal ticket links, customer names and any figure the manager has not approved for publication.
```

### Save the drafts

```text
Use Tareno to confirm my X account [handle] and my LinkedIn profile [name], then save the two approved texts as separate drafts. Return each draft ID with its text, and request no time until I give you the day and the time zone.
```

### Request the slots

```text
Take the drafts [draft IDs] and ask Tareno to schedule them for [Friday 16:00] in [Europe/London]. Return the action IDs and remind me that each request still waits for my approval in the Tareno dashboard.
```

## Guides by network

- LinkedIn: [Can a Paperclip agent draft a manager's monthly retrospective for LinkedIn?](https://tareno.co/agents/paperclip/linkedin) ([Markdown](https://tareno.co/agents/paperclip/linkedin.md))
- X: [Can a Paperclip agent turn a week of closed issues into an X progress post?](https://tareno.co/agents/paperclip/x) ([Markdown](https://tareno.co/agents/paperclip/x.md))

## Good to know before you start

- Tareno has not tested a Paperclip setup end to end. The steps follow Paperclip's documentation for the Claude Code and Hermes adapters.
- Paperclip has no MCP command of its own. A server configured in Claude Code's user scope or in Hermes's host-wide YAML is visible to every agent run by that operating-system user, not to one agent only.
- A heartbeat runs unattended, so OAuth must be completed beforehand as the same user, and an expired token will fail the run until someone signs in again.
- Third-party projects named paperclip-mcp expose Paperclip's own API to a client. They do not connect Paperclip agents to Tareno, and none is confirmed as official.
- Scheduling, publishing, edits to scheduled posts and deletions create approval actions in Tareno. A Paperclip agent can request them but cannot approve them.

## Paperclip compared with Hermes Agent and OpenClaw

Hermes Agent and OpenClaw are agents with published Tareno skills. Paperclip sits above agents like them and decides when they run and within which budget. If you need a single agent that drafts, start with those pages. If you run several agents with goals and a manager, Paperclip is the layer that schedules them, and the Tareno connection stays in each runtime.

## Frequently asked questions

### Does Paperclip have its own Tareno connector?

No. Paperclip's documentation says it has no MCP command, because the control plane launches the runtime and the runtime speaks MCP. Tareno is therefore added in the Claude Code or Hermes configuration that the agent uses, and Paperclip only schedules the runs.

### Which Paperclip adapters are documented for MCP?

Claude Code and Hermes. The guide expects Codex, Cursor Local, Gemini CLI, OpenCode and Pi to inherit MCP from their own CLIs but says that path is not covered yet. Tareno makes no claim about those adapters.

### How does sign-in work for unattended heartbeats?

You complete the OAuth flow once in an interactive session as the same operating-system user that runs the heartbeat. Tokens persist and refresh automatically. If the user differs, the heartbeat will not find the credential and the tool list will be empty.

### Can I scope Tareno to one agent?

Only partly. A project-level .mcp.json applies to agents working in that directory, while user scope and Hermes's YAML apply to everything run by that operating-system user. Because Tareno needs OAuth, user scope is the documented choice for remote servers.

### Is the paperclip-mcp project the same thing?

No. Community projects with that name expose Paperclip's REST API to clients like Claude Code, so a person can manage Paperclip from a chat. They do not give Paperclip agents access to Tareno, and none is confirmed as official.

### Has Tareno tested this connection?

Not end to end. The steps come from Paperclip's own guide and Tareno's documented Claude Code and Hermes setups. Run claude mcp get tareno and a read-only call from the agent's directory before you let a heartbeat save drafts.

### Can an agent publish a post without my approval?

No. A scheduling or publishing request becomes an approval action in Tareno, and nothing reaches X or LinkedIn until you approve it in the dashboard. The agent can only report that the action is pending and read its status afterwards.

### Which Tareno tools will the agent call first?

Usually `list_accounts` to confirm the destination, then `create_drafts` to save the text. `schedule_posts` comes last and returns an approval action instead of publishing anything, and `get_action_status` tells you what became of the request.

## Tareno tools used

- `list_accounts`
- `create_drafts`
- `schedule_posts`
- `get_action_status`

Approval model: Tareno saves drafts directly. Scheduling, publishing, edits to scheduled posts and deletions become pending actions that a person approves in the Tareno dashboard; no agent can approve them.

## Related

- [A workstream update on X](https://tareno.co/agents/paperclip/x)
- [A manager's retrospective on LinkedIn](https://tareno.co/agents/paperclip/linkedin)
- [Remote MCP setup](https://tareno.co/docs/mcp)
- [Claude Code × Tareno](https://tareno.co/agents/claude-code)
- [Hermes Agent × Tareno](https://tareno.co/agents/hermes-agent)
- [Approval workflows](https://tareno.co/features/approval-workflows)