A compatible assistant
Peripheral works with ChatGPT, Claude, Grok and Perplexity, and any client that supports remote MCP over Streamable HTTP with OAuth sign-in. Check that your account permits custom MCP connections.
Get startedConnect in ~2 min
Connecting takes about 2 minutes. Importing depends on how much you bring.
Already connected? Go to the promptsOne minute. Sign up with your work email or single sign-on. You only do this once.
In Claude: Customize → Connectors
Your assistant acts with your permissions.
Works on every plan (Free allows one custom connector). On Team or Enterprise, your admin must allow custom connectors. Menus change often. Last checked Nov 2026.
Set up? Browse the prompt library →
One prompt does both: it brings in your existing work, then sets up your agents to keep The Record current. You paste it once and finish in your assistant.
Add one or several. Each gets its own space. This stays in your browser.
2 · Already working on this with your assistant?
3 · Sources to keep current Choose at least one
Be specific where possible. Your assistant will resolve access and scope before configuring updates.
5 · How often
These are preferences. Your assistant checks what it can actually run.
Paste this into the conversation where you’ve been working on this company or project.
Use the connected Peripheral MCP tools to set up Peripheral in two parts, in this order, in this conversation. Finish Part 1 before starting Part 2. PART 1 — Build the initial record. Work on [company]. 1. Find the space. Look for an existing Peripheral space for [company] first. If none exists, ask me to confirm before creating one. 2. Gather what you can access. Use this conversation, any earlier history you can reach, files I've shared and connected sources. Don't assume a folder named after the company exists. If the scope of a source is unclear, confirm it with me before importing. If you can't access earlier conversations, say so and work with what's available. 3. Ask only what's essential. Check what you already know, then send me the questions you still can't answer in one short message: my role, the decision or work I'm preparing, the sources that matter most, and anything to leave out. 4. Check what Peripheral already holds and avoid duplicates. 5. Package and upload. Add original files where possible, preserving their sources and dates. Put context that exists only in conversation into a dated note named "[company] — Initial context — [date].md". Separate documented facts, information I supplied, your interpretations, assumptions and open questions. Upload to the correct space, confirm each upload and check processing status. 6. Report honestly. Do not stop at a summary in chat. List what Peripheral received, what is still processing, what you skipped and what failed. If you can't create or transfer files, say so and give me the prepared note without claiming it was added. PART 2 — Set up ongoing updates. Start this as soon as you've reported Part 1. It is setup, not another one-off import. 1. Verify the target. Confirm the Peripheral space for [company]. Don't create one without asking me. 2. Resolve the sources. I'd like updates from: sources we agree on (ask me). Explicit source restrictions take precedence over broad category selections. If they contradict a selected category, clarify with me before proceeding; do not request or ingest from the broader category. Tell me exactly which source locations or records you can access, and which you can't. Only use sources we agree on here. 3. Agree the scope. Confirm what's in, what's excluded and the timing. My preference is whatever schedule or trigger you can support. 4. Check capability. Tell me whether your environment can run on its own on that schedule or trigger AND upload to Peripheral each time. Name the specific missing capability if not. 5. Configure it. If supported, set up the automation within my authorization. If not, explain what's missing and give me a short on-demand prompt I can run instead. 6. On each run, add only new or changed relevant material, check for duplicates, preserve sources and dates, and keep evidence separate from interpretation. Verify each transfer and report failures. 7. Use only these receipt states: Needs setup; Scheduled — awaiting first run; Active — update verified; Manual only; Needs attention. Report Active only when recurring execution is configured AND at least one recurring run has completed successfully. A successful one-off import is not evidence of recurring automation. Never say updates are active just because I accepted this. Whenever the status is Needs setup or Needs attention, write it as 'Needs setup — [exact missing item]' or 'Needs attention — [exact blocker]' (for example 'Needs setup — folder path not confirmed'), and ask me the single question needed to resolve it. Never use vague summaries like 'sources not identified or connected'. Name each item individually: for every source I selected, say which service it is (e.g. Google Drive, Slack), whether you can access it right now, and the exact folder, label, channel or record you still need from me. List each missing item on its own line with the action to fix it, e.g. 'Files and drives — Google Drive not connected in this assistant: connect it under connectors' or 'Slack — channel not named: tell me which channel, e.g. #deal-acme'. 8. Return a setup receipt: destination space, sources, agent or service responsible, schedule or trigger, status (Needs setup, Scheduled — awaiting first run, Active — update verified, Manual only or Needs attention, with the exact missing item or blocker when relevant), next step needed (the one concrete action or answer required from me, or 'None'), last successful update, next scheduled check and any failures needing attention. Finish by suggesting one useful first question I could ask about [company], based on the material received.
States: Needs setup · Scheduled — awaiting first run · Active — update verified · Manual only · Needs attention. Active requires recurring execution and a verified successful run; a one-off import is not enough. Needs setup and Needs attention always name what’s missing and the next step, e.g. “Needs setup — folder path not confirmed”.
Change or pause the schedule in the agent or service that runs it, then ask for a fresh receipt to check. Disconnecting Peripheral doesn’t delete a schedule set up elsewhere.
✓Signed in as you@acme.com · Peripheral connected
Shall I import these?
Setup receipt
Active — first update verified
Daily update: 4 new files and 12 messages added. The FY target changed in the new board pack. No failures.Illustrative example. Not your account’s data.
Some workspaces need an admin to allow custom connectors.
Peripheral works with ChatGPT, Claude, Grok and Perplexity, and any client that supports remote MCP over Streamable HTTP with OAuth sign-in. Check that your account permits custom MCP connections.
Add the server in your client’s MCP settings and sign in to Peripheral. The connection acts as the signed-in user, with that user’s permissions.
Select the company’s space and import what you already have. Then set up your agents with a supported schedule or trigger so new material keeps arriving.
Choose the decision, the output, and who reviews it. Start with a cited briefing for an existing diligence, strategy review, board meeting or monitoring cycle.
Add the endpoint as a remote HTTP MCP server, complete Peripheral’s OAuth sign-in, and verify the connected account before selecting a company.
Development environment. Confirm the correct environment and access with your team.
SOC 2 Type II. Encrypted in transit and at rest. Not used to train external models. Trust Center ↗