MCP skills catalog

When you connect Clment to Claude Desktop (see Connecting Claude Desktop), Claude can call any of the skills below. Each skill is a tool the agent picks based on your request — you don’t choose the skill, just describe what you want.

Read tools

Information retrieval. Always free; never modify your data.

• search_contracts — find contracts by title, party, status, type, tags, or natural language. Supports createdAfter / createdBefore date filters.
• get_contract — fetch a single contract by CLM-N. Returns metadata, key dates, status, the parties, and a summary.
• get_contract_text — fetch the full extracted text of a contract. Useful when the agent needs to reason over specific clause wording.
• get_contract_file — issue a signed download URL for the original document (PDF/DOCX).
• get_review — fetch a review by REV-N. Returns the findings, verdicts, comments, redline-inclusion flags.
• get_review_report — produce the formatted Word (DOCX) review report (the same one you’d export from the UI).
• list_playbooks — list available playbooks. Returns name, description, tags, status (draft/published).
• get_playbook — fetch the full clauses + recommended language of a specific playbook.
• list_my_reviews — your reviews (assigned, in-progress, completed).
• list_reviews — every review in the org.
• list_my_mentions — places where you’ve been @-mentioned.
• list_activity — unified activity feed for a contract (uploads, revisions, reviews, comments).
• list_key_dates — upcoming key dates across the portfolio.
• get_insights — portfolio-level analytics (risk distribution, expiry pipeline, top counterparties).
• get_org_summary — a snapshot of your workspace: the org name, the active users and their roles, and how many contracts, reviews, and playbooks you have.
• get_resource_links — fetch UI deep-links for a contract or review (handy when the agent needs to send the user to the right page).
• ask_question — natural-language portfolio Q&A with citations. The same RAG that powers the in-app AI Assistant.
• get_help_articles — ask the help center a product question. Same RAG as this page’s chat widget.

Write tools

Actions that modify state. Each one logs to the audit trail with attribution to your name + email (passed through the MCP token).

Upload + ingestion

• request_contract_upload — start a multi-step upload, returns an upload-URL the agent uses to push the file.
• get_upload_status — check whether an upload has finished ingesting + been assigned a CLM-N.
• upload_contract_from_url — single-step upload: agent supplies a public URL, Clment fetches + ingests.
• convert_pdf_to_word — convert a PDF contract version to a Word (.docx) document and save it as a new version on the contract. Idempotent; charged per 50-page block.
• convert_pdf_url_to_word — convert a PDF at a public URL to Word and return a short-lived download link. Doesn’t create a contract — for one-off conversions.
• attach_referenced_document — fetch a URL the extraction pass flagged on a contract (its pendingReferencedDocuments) and save it as a Supporting attachment. Free.

Review lifecycle

• start_review — kick off a review of a contract with a chosen playbook. Optionally pass a review strategy (explore / negotiation / high_priority / strict) to set how assertive the review is and what lands in the redline; defaults to the playbook’s default, or Explore.
• assign_review — assign an in-progress review to a teammate.
• add_finding_comment — add a comment to a specific finding.
• set_finding_decision — the unified 4-control update: verdict + comment + include-in-redline + redline instruction. See Understanding finding verdicts.
• generate_redline — produce the tracked-changes Word document from the review’s current finding states.
• sign_off_review — lock the review.

Metadata

• set_contract_status — change a contract’s status (draft, active, expired, terminated, superseded).
• set_contract_risk — set risk level (unknown, low, medium, high, critical).
• update_contract_tags — add / remove tags.
• update_reminder — change reminder settings on a key date.

Read vs write — what each MCP scope grants

When you authorise Clment in Claude Desktop, you’ll see a consent screen listing the skills the connector is requesting. Two scopes:

• read — every read tool above. Free; no credits consumed. Cannot modify any data.
• read_write — read tools + write tools. Credit-metered actions (start_review, generate_redline, ask_question) still consume credits as if you’d done them in the UI.

You can revoke the connector at any time from Settings → Connect AI Agent in the Clment web app. That immediately invalidates the token; the next API call from Claude Desktop returns 401.

Examples of what to ask

• “List my contracts expiring in the next 60 days, sorted by risk.”
• “For REV-12, accept all findings the AI recommends as ‘fail’ and reject the rest.”
• “Upload [URL] and start a review with our standard vendor playbook.”
• “Show me everything about CLM-42 — text, key dates, recent activity.”
• “What does Clment cost?” — uses get_help_articles to answer from this help centre.

See also

• Connecting Claude Desktop
• MCP troubleshooting
• Understanding finding verdicts