> ## Documentation Index > Fetch the complete documentation index at: https://docs.chatropic.com/llms.txt > Use this file to discover all available pages before exploring further. # Available MCP tools > The complete Chatropic MCP tool catalog, permissions, inputs, and result limits. Chatropic provides **16 MCP tools**. Your AI client discovers only the tools covered by your approved permissions. Describe what you want in natural language; the client supplies the tool inputs. The reference below also helps you check exactly what a tool can do. [Connect your client](https://docs.chatropic.com/user-guides/chatropic-mcp/connect.md) before trying these tools. For complete examples, see [Workflows](https://docs.chatropic.com/user-guides/chatropic-mcp/workflows.md). ## Agents Requires `agents:read`. | Tool | What it does | Inputs | | - | - | - | | `list_agents` | Lists agents in your workspace with IDs and sanitized configuration summaries. Some text fields may be redacted or withheld. | Optional `limit`, `cursor` | | `get_agent_config` | Returns a limited configuration summary for one agent, such as enabled state, channels, model, and tool names when available. It does not expose instructions or secrets. | Required `agent_id` | These are inspection tools. They cannot change an agent's instructions, channels, model, or tool grants. Configuration summaries currently read stored release data, so they may not reflect the latest saved Builder configuration; check Builder when those details matter. ## Support conversations Requires `conversations:read`. These tools cover web conversations; they are not an export of every channel's inbox. | Tool | What it does | Inputs | | - | - | - | | `search_conversations` | Finds conversations by message text and date window. Returns IDs, agent IDs, statuses, and timestamps rather than full transcripts. | Optional `query`, `days`, `limit`, `cursor` | | `get_conversation` | Reads a conversation's customer-facing user and assistant messages, with redaction. | Required `conversation_id`; optional `limit`, `cursor` | An empty query lists conversations within the selected window. The default window is 30 days; `days` supports 1–90. Use `get_conversation` on the returned IDs to review messages. These tools cannot reply to customers, close conversations, or change assignments. ## Support analytics Requires `analytics:read`. | Tool | What it does | Inputs | | - | - | - | | `get_support_metrics` | Returns aggregate web conversation and resolved counts for the last 30 completed days. Small groups are suppressed. | None | The window is fixed. This tool does not accept custom dates or provide a full analytics dashboard. A suppressed or unavailable value is not zero. ## Knowledge and content gaps Requires `knowledge:read`. | Tool | What it does | Inputs | | - | - | - | | `search_knowledge` | Searches titles and article bodies across non-deleted knowledge, including private drafts and agent-disabled articles. An empty query lists available content. | Optional `query`, `limit`, `cursor` | | `get_knowledge_article` | Reads one article's title, current draft text or published fallback, published text, status, and agent-enabled state. | Required `article_id` | | `list_knowledge_gaps` | Lists already-detected content gaps with titles, summaries, status, and linked article IDs when available. | Optional `query`, `status`, `days`, `limit`, `cursor` | Article and gap text is returned verbatim. Search results can include unpublished content; inclusion in a result does not mean an article is public or enabled for an agent. For gaps, `status` can be `all`, `open`, `addressed`, or `dismissed` and defaults to `all`. The default window is 45 days; `days` supports 1–365. Listing gaps does not run new analysis or mark gaps addressed. Manage those actions in [Content gaps](https://docs.chatropic.com/user-guides/analytics/content-gaps.md). ## Drafting and editing documentation Requires `knowledge:write`. | Tool | What it does | Inputs | | - | - | - | | `get_documentation_skill` | Returns Chatropic's documentation-writing guide to help the AI compare product information with existing articles and prepare help content. | None | | `list_knowledge_drafts` | Finds articles and their draft/published versions, revision numbers, and source references. It can include articles that already have a published version. | Optional `query`, `article_id`, `limit`, `offset` | | `create_knowledge_draft` | Creates a private draft with agent access disabled. It does not publish the article. | Required `tenant_id`, `title`, `markdown`, `request_key`; optional `folder_id`, `provenance` | | `update_knowledge_draft` | Replaces an article's draft title and Markdown only if the expected revision still matches. It does not publish the update. | Required `article_id`, `expected_revision`, `title`, `markdown`; optional `provenance` | For creation, `tenant_id` is the target agent ID within the connected workspace. Obtain the correct ID from Chatropic or `list_agents` if agent reading is authorized. The server checks that it belongs to your workspace. Draft listings use `offset` rather than a cursor: the default page size is 25, the maximum is 50, and the maximum offset is 10,000. Inventory results shorten draft and published Markdown to 2,000 characters each. Supply `article_id` to retrieve the full versions before editing. An update supplies the **complete replacement draft**, not a patch. The AI should read the full article, preserve sections outside your requested change, and send its current `revision` as `expected_revision`. If another editor has changed it, read again before retrying. Titles support up to 300 characters and Markdown up to 100,000 characters. `provenance` records source references used for the documentation. On creation, use a new `request_key` for each new article; reuse it only to retry an identical request. This helps prevent duplicate drafts after a connection interruption. ## Publishing an article Requires `knowledge:publish`, separately from draft editing. | Tool | What it does | Inputs | | - | - | - | | `publish_knowledge_article` | Publishes the specified current article revision. Rejects an outdated revision so the client cannot silently publish a different edit. | Required `article_id`, `expected_revision` | Review the full draft before publishing. Publishing does not automatically enable agent access or opt the article into the public Help Center. Existing visibility settings still apply, so publishing changes to an already-visible article can update what readers see. Review distribution settings in Chatropic's Knowledge interface. This publishes a **knowledge article**; it is not an agent deployment or a requirement for connecting MCP. ## Billing and usage Requires `billing:read`. | Tool | What it does | Inputs | | - | - | - | | `get_billing_summary` | Reads available plan, seat, subscription, trial, and billing-period information. Payment details are excluded. | None | | `get_billing_usage` | Reads stored usage counts, allowances, and remaining credits when available. | None | | `list_billing_invoices` | Lists invoice or transaction summaries, up to the latest 100 records. Payment details and links are excluded. | Optional `limit`, `cursor` | Availability depends on the billing provider and recorded data. Missing values mean unavailable, not zero usage or an inactive subscription. These tools cannot change a plan, purchase credits, update payment details, or issue refunds. ## Results and limits Most list tools accept `limit` from 1–100, with a default of 25, and return `next_cursor` when more results are available. Continue with that cursor using the same tool and search parameters. Do not reuse it with another query or connection. Draft listings use the separate offset pagination described above. Successful results include a `request_id` for tracing and a `redacted` indicator. Knowledge and documentation tools return unredacted content; other categories apply sanitization or redaction. An empty list can mean no matching or accessible records, rather than a server failure. The server currently allows up to 60 calls per minute per grant, 300 per minute per workspace, and five simultaneous calls per grant. These are execution limits, separate from selecting permission categories. Calls have a 15-second execution timeout and a response-size limit of approximately 250 KB. Narrow the query or reduce the page size if results are too large. The server does not provide tools for deleting articles, executing arbitrary SQL or code, managing users, changing workspace permissions, or changing agent configuration. Use the Chatropic dashboard for actions outside this catalog. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.