To use 1PC with Mistral's browser assistant, connect https://use1pc.com/mcp through a custom MCP connector in Vibe Work, authorize your 1PC account and explicitly enter Product Owner. Start by reading product state, then verify one scoped save in a fresh conversation.
Mistral now presents Vibe as the successor to Le Chat. This guide follows the Work connector interface documented on 4 October 2026. It describes a setup to verify in your account; it is not a record of a completed live 1PC connection.
Choose Mistral Work for the browser connection
A Mistral model and a Mistral application are different parts of the setup. The application needs to discover and call 1PC's remote tools, handle authorization and show the result. Choose the surface before following a configuration example:
| Surface | Role in this guide |
|---|---|
| Vibe Work | The selected browser route: add a custom MCP connector and complete OAuth. |
| Vibe Code CLI | A separate configuration. Its MCP documentation currently says OAuth servers are unsupported; do not reuse the browser procedure there. |
| Mistral Studio/API | A developer route with its own connector lifecycle. The Studio documentation labels that interface Public Preview. |
You need a 1PC account, a completed founder profile and access to the product you intend to use. The MCP overview explains the shared product layer; the MCP reference supplies authentication and tool details. Your Mistral subscription does not change your 1PC plan.
Add the custom connector and authorize 1PC
Mistral's custom MCP connector documentation describes an administrator-managed setup with OAuth 2.1 dynamic registration. Use these fields for 1PC:
- Open Connectors, select + Add Connector, then Custom MCP Connector.
- Set the connector name to
onepc, the server URL tohttps://use1pc.com/mcpand an optional description such as “Saved product knowledge, decisions and progress.” - Select Connect and complete the detected OAuth flow in your browser.
- Inspect the 1PC account and requested access before granting consent.
1PC exposes remote Streamable HTTP. Use the full /mcp endpoint, with automatic client registration; it does not advertise CIMD support. Reading uses knowledge:read. A save requires both knowledge:read and knowledge:write. The host handles credentials; your website login cookie is not the MCP credential.
To enable the connector for a task, Mistral documents opening + or /, then choosing Tools. See its connector usage guide. Confirm that onepc is enabled before requesting a product read.
Load Product Owner before asking for deliverables
Connecting the server makes tools available. It does not install 1PC's native Product Owner skill into Mistral. Make the entry procedure explicit for the first read and repeat it for later product tasks unless your host retains an appropriate instruction:
Use the onepc connector for this request. First read
get_workspace_context and load get_skill_instructions with
skill_id "product-owner" and phase "entry".
List my accessible products and let me select the relevant product.
Load get_product_context with its product_id.
Explain the current facts, accepted decisions and unfinished work.
If a summary is truncated, retrieve the remaining relevant records.
This is a read-only check. Do not save, rerun earlier work or create
a product.
The assistant should use the server's actual tool schemas and selected product ID. Inspect its tool activity: a plausible summary written without a product read has not established the connection.
Product Owner discovers the paginated list_product_skills catalog and loads selected guidance through get_skill_instructions. Every User has the available skill catalog and private workspace skills. Accepted private MCP calls consume prepaid credit; loading guidance does not execute work or install a native skill.
Try a feature-claim review before a larger campaign
A useful first task is to compare a draft's claims with the saved product scope. It tests whether the assistant can distinguish what is true, what is planned and what still needs evidence. It also connects product and marketing work without starting an entire launch.
Consider Follow-up Board, a fictional tool for freelance designers tracking unanswered proposals. The example below is a review framework, not evidence about a real product:
| Draft claim | What the assistant must check | Review outcome |
|---|---|---|
| “Keep unanswered proposals in one board.” | Current scope and any saved release evidence. | Supported only if those sources establish availability. |
| “We send every follow-up automatically.” | Whether delivery is implemented, planned or excluded. | Correct or qualify the claim when the scope does not support it. |
| “Win twice as many clients.” | A dated measurement with a meaningful comparison. | Mark as unsupported if that evidence is absent. |
Apply a published private launch-review skill from your workspace. Publish the method before asking the assistant to use it. Product Owner should report a missing skill instead of silently substituting another method.
Enter Product Owner again for the selected product. Discover my
published launch-review skill and load its current work guidance.
Use its required product context to review the draft I provide.
Identify each claim as supported, planned or unknown, with its source.
Suggest the smallest corrections and one next action. Keep missing
evidence explicit. Review only; do not save or publish yet.
If you do not have that method, begin with the read-only state check. The shared feature-launch marketing guide explains how to turn an accepted feature into a brief and selected assets; it owns that method across providers.
Save the reviewed result and retrieve it in a fresh task
Once the result is useful, request a bounded save:
Save this reviewed claim assessment and its open evidence gaps as
scoped Knowledge with an explicit department, reference role and stable key.
Prepare and self-review the update before committing its exact revision.
Read the current file and revision first; preserve valid information.
Show the save receipt, then read back the changed file. Leave the
marketing claims proposed; do not publish or contact anyone.
The assistant should return the actual path and revision. If the response is uncertain, read that destination before retrying. A second blind save can overwrite a valid result.
Open a fresh Work conversation, enable onepc and enter Product Owner again. Select the same product, provide the returned path and ask it to read the assessment without rerunning the review. Verify that the saved corrections and open gaps are present. That checks stored continuity; it does not promise identical reasoning from every model.
The cross-tool context guide explains this boundary. Unsaved messages, running host tasks and temporary attachments remain outside 1PC unless their useful results are explicitly saved.
Separate connection, access and task failures
Work's current MCP documentation notes limits around resources, automatic prompt templates and dynamic discovery. This guide therefore uses explicit tool requests rather than assuming the host imports 1PC prompts. If tools change and your connector does not refresh them, recheck the current configuration before continuing.
- The connector cannot be added: check your Mistral administrator access and the full HTTPS endpoint.
- Authorization fails: inspect OAuth registration, consent and callback errors using the 1PC authentication reference. Do not invent client IDs or secrets.
- Reads work, saves fail: check write consent, product membership and current 1PC access separately.
- The review skill is missing: check that it is published, active and in the caller's own workspace; a shared product does not share private skill libraries.
- The next conversation lacks the result: confirm the account, selected product and actual saved path. A statement that it “remembers” is not a read-back.
Record the interface, date and observed read/save/read-back when testing. Once this loop works, use the existing weekly product review or launch guide for the next task. 1PC stores authorized results; research, asset generation and any external action still depend on available host tools and your request.
Check your Mistral-to-1PC setup.
Use the MCP reference to verify authentication, tool access and a saved continuation.
Open the MCP reference ↗