cd /news/ai-agents/five-crm-workflows-to-hand-to-claude · home › topics › ai-agents › article
[ARTICLE · art-143761] src=relaticle.com ↗ pub= topic=ai-agents verified=true sentiment=· neutral

Five CRM workflows to hand to Claude

Relaticle's agent-native CRM registers 39 MCP tools as of September 22, 2026, and the company published five paste-ready Claude workflows — a read-only Monday pipeline review, meeting-notes-to-CRM, inbox-to-CRM data entry, a data hygiene sweep, and an end-of-week digest. Relaticle warns that Claude connected through MCP commits writes directly, unlike the built-in chat assistant that routes proposed writes through human approval cards, so users must inspect proposed records and confirm each deletion before saying yes.

by read9 min views1 publishedOct 2, 2026
Five CRM workflows to hand to Claude
Image: Relaticle (auto-discovered)

Claude can take useful CRM work off your plate, but only if you give it bounded jobs and treat writes differently from reads. Start with a read-only review or digest. When you add writes, make Claude show its proposed changes and wait for your go-ahead, because Relaticle MCP writes commit directly. The approval gate is your conversation, not the app.

Relaticle registers 39 MCP tools, checked September 22, 2026. They cover CRM records, workspace context, and uploads. That surface is enough for repeatable work, not just one-off questions. The companion post, Let Claude Code run your sales pipeline, showed the raw sessions. This is the recipe card: five prompts you can paste, what each one changes, and where to be careful.

The safety distinction matters, and it comes from Relaticle's agent-native design: the built-in chat assistant turns proposed writes into approval cards that a human can approve or discard. Claude connected through MCP does not. Its tool writes commit directly. Asking Claude to list its plan and wait is not an application-level safeguard. It is a useful conversational control, and you still need to inspect the proposed records before you say yes.

#

Workflow Mode
Monday pipeline review Read-only
Meeting notes to CRM Write
Inbox-to-CRM data entry Write
Data hygiene sweep Read, then write
End-of-week digest Read-only
Workflow Useful for
Monday pipeline review Seeing the week's opportunities and stale work
Meeting notes to CRM Turning a conversation into notes and follow-ups
Inbox-to-CRM data entry Capturing leads from signatures or a lead list
Data hygiene sweep Finding likely duplicates and stale records
End-of-week digest Planning the next week
Workflow Write boundary
Monday pipeline review None
Meeting notes to CRM Review the proposed records before go-ahead
Inbox-to-CRM data entry Review names, companies, and field values before go-ahead
Data hygiene sweep Confirm each deletion explicitly
End-of-week digest None

For setup, start with the MCP guide and the walkthrough for connecting Claude or ChatGPT. The workflows below assume Claude can already reach your Relaticle workspace.

#

This is the easiest place to begin because it does not write anything. Ask Claude to list opportunities, group them by stage, and call out the ones that look stale. The point is not to let a model decide what to do with a deal. The point is to turn the first review of the week into a short, structured queue for your own judgment.

Paste this:

Review our current opportunities for Monday planning. Follow pagination until has_more is false before computing totals.

List the opportunities, group them by stage, and identify any that appear stale based on the CRM record details. For each flagged opportunity, include the company and explain briefly why it needs attention.

Do not create, update, or delete anything. End with the smallest practical list of follow-ups for me to review.

Tools it leans on: list-opportunities, get-opportunity, and company lookups. (Tool names here are shortened; most wire names carry a -tool suffix, e.g. list-opportunities-tool, while search, fetch, upload-file, and create-upload-url omit that suffix.)

Gotcha: "Stale" needs to come from the record details you can inspect. Do not ask Claude to invent a definition from outside the CRM. If your team has a specific stale-deal rule, put that rule in the prompt before you run it.

The output is most useful when it gives you a starting list, not a verdict. A deal can look inactive in the CRM while an important conversation is happening elsewhere. That is why this workflow stays read-only.

#

Meeting notes are a good write workflow because the source material is in front of you. Paste the raw notes, then have Claude turn them into a note, attach it to the relevant company and person, and propose follow-up tasks with assignees. This reduces transcription work while keeping you in control of where the record lands.

Paste this, then replace the bracketed section with your notes:

Turn the raw meeting notes below into CRM updates.

[PASTE RAW MEETING NOTES]

Find the matching company and person. Create one concise CRM note from the meeting, attach it to that company and person, and propose follow-up tasks with assignees when the notes identify an owner.

Before making any write, list what you plan to change and wait for my go-ahead. Do not create, attach, or assign anything until I explicitly approve the plan.

Tools it leans on: who-ami-tool, create-note-tool, attach-note-to-entities-tool, create-task-tool, and attach-task-to-entities-tool.

Gotcha: Notes often contain ambiguity. A person's name may not identify one record, and an action item may not name an assignee. Do not force a match. Use who-ami-tool to discover workspace members before proposing assignees. Resolve uncertain matches before approving writes.

This prompt deliberately asks for a plan before any tool call that writes. Claude has no approval card in this path. Once you approve, the MCP writes commit directly.

#

Signature blocks and lead lists create small backlogs that are boring to enter one by one. Claude can create people and companies from the text you paste. It can also fill custom fields, which are readable and writable over MCP without per-field integration code (how that works).

Paste this:

Add the leads in the text below to our CRM.

[PASTE A SIGNATURE BLOCK OR LEAD LIST]

For each lead, identify the person and company information present in the text. Create the people and companies, and fill relevant custom fields only when the supplied text supports the value. Do not guess missing values.

Before making any write, list what you plan to change and wait for my go-ahead. Show the people, companies, and custom field values you intend to create. Do not create anything until I explicitly approve the plan.

Tools it leans on: create-people and create-company.

Gotcha: A signature block is not a complete profile. Treat absent information as absent. The same rule applies to custom fields: use a value only when it appears in the pasted source. If you need a field that does not exist, do not ask Claude to improvise one. Review the available choices in the custom field types guide. FileUpload and MarkdownEditor are retired types.

The plan is also where you catch duplicates before they enter the CRM. If the pasted company name resembles an existing company, ask Claude to call that out rather than treating a new record as automatic.

#

This is the workflow to approach with the most caution. First use Claude to find likely duplicates and stale records. Then make it present the list. Only after you inspect the candidates should you authorize any deletion. Read and write are separate phases here, on purpose.

Paste this:

Run a data hygiene review of our CRM.

Use the available list and search tools to find likely duplicate records and records that appear stale from their CRM details. Present a review list with the record type, record name, reason it was flagged, and the exact deletion you would recommend, if any.

Do not delete anything during this review. After I review the list, list what you plan to change and wait for my go-ahead. Delete only the specific records I explicitly confirm. Do not infer approval for similar records.

Tools it leans on: list tools, search, and the relevant delete-* tool.

Gotcha: "Likely duplicate" is a review category, not proof. Similar names can represent different people or companies. A stale record may still matter. This is where the "wait for my go-ahead" pattern is non-negotiable because deletion is direct in Claude's MCP connection.

Keep the confirmation narrow. Approve named records, not a broad instruction such as "clean up the rest." That makes the proposed action easier to inspect. Server permissions and client approval settings still determine what the connection can execute.

#

The end-of-week digest is another read-only workflow. It gives you a compact view of what changed, which tasks are due next week, and which deals need attention. Use it to prepare a handoff, plan the next week, or notice work that did not move.

Paste this:

Prepare an end-of-week CRM digest.

Read the activity feed for changes this week. List tasks due next week and opportunities needing attention. Follow pagination before summarizing totals. Use the CRM records to support each item. Group the result into: changes this week, tasks due next week, and deals needing attention.

Do not create, update, or delete anything. Keep the digest concise enough for planning.

Tools it leans on: list-activity-tool, list-tasks-tool, list-opportunities-tool, and search.

Gotcha: A digest only reflects what made it into the CRM. It cannot include work that stayed in an inbox, a calendar, or a private note. Treat it as a CRM status report, not a complete account of the week.

#

The common thread is simple: ask Claude to retrieve and organize information first. Start with the Monday pipeline review or the end-of-week digest. Add write workflows when you trust the loop, the source text, and your own review step.

For every write workflow, make the prompt say, "list what you plan to change and wait for my go-ahead." That phrase creates a deliberate , but it does not replace a system approval gate. In Claude, MCP writes commit directly. Use the to inspect the records, resolve ambiguity, and approve only the exact changes you want.

When you are ready to connect the workspace, follow How to connect Claude to your CRM with MCP. Then start read-only, and add writes one bounded workflow at a time.

    [Guides](https://relaticle.com/blog/backups-upgrades-and-security-for-a-self-hosted-crm)

Backups, upgrades, and security for a self-hosted CRM #

A practical operating playbook for backing up, upgrading, and securing a self-hosted CRM without treating customer data casually.

    [Guides](https://relaticle.com/blog/the-real-cost-of-self-hosting-a-crm)

The Real Cost of Self-Hosting a CRM #

Self-hosting a CRM can be cheap, but it is not free. Here is the infrastructure, time, and SaaS comparison math.

    [Guides](https://relaticle.com/blog/migrating-from-hubspot-or-attio-to-a-self-hosted-crm)

Migrating from HubSpot or Attio to a Self-Hosted CRM #

Plan a HubSpot or Attio migration: export records and notes, map relationships, test imports, and identify the history and files needing separate work.

── more in #ai-agents 4 stories · sorted by recency
── more on @relaticle 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/five-crm-workflows-t…] indexed:0 read:9min 2026-10-02 · —