Carta

What the Carta connector does

The Carta MCP server connects private-market equity records to Claude, ChatGPT and any MCP client. It answers questions about cap table ownership, 409A valuations, SAFEs, vesting schedules and fund performance against the ledger a company or fund already runs on. Anthropic's directory labels its permissions Read. Three independent checks say that label is incomplete.

Carta tools (9)

• welcome
• discover
• fetch
• list_contexts
• set_context
• list_accounts
• get_current_user
• cap_table_chart
• cap_table_healthcheck

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

• The directory's "Read" label is incomplete. Three readwrite_ OAuth scopes are advertised, Carta's CRM guide states the plugin writes, and a published Carta skill issues securities. Two of the three plugins are genuinely read-only; the connector as a whole is not.
• Nine is not the tool count. Three of the nine listed names are dispatchers over a command registry, and Carta's own skills call tools — call_tool, search_tools, mutate — that the directory never lists. The true surface is unenumerable without authenticating.
• The server is stateful and the active context is not readable. set_context selects a fund administration firm for later calls, the model may change it at any turn, and no tool reports which firm is currently active. A wrong-context answer is possible and would not announce itself.
• Carta documents nothing about contexts. None of the four published guides mentions contexts, context switching, or a login spanning several firms. The discipline exists only inside plugin skill files, and does not reach a client connecting with the bare URL.
• cap_table_healthcheck provenance is unresolved. The tool appears in Anthropic's snapshot and in none of Carta's documentation or public source. Whether its output is a retrieved Carta report or a generated assessment is unknown, and an assessment of a company's equity structure is a materially different artefact from a report.
• welcome and cap_table_chart are likewise undocumented by name. Carta's skills describe a once-per-connection handshake that caches identity and firm data for 24 hours, but no vendor documentation ties that behaviour to the listed tool.
• No tool is annotated. The endpoint returned 401 to an anonymous request on 2026-08-19, so no inputSchema, readOnlyHint or destructiveHint was readable for any of the nine.
• The read surface reaches named individuals' equity records. Stakeholder lists, grants with vesting schedules, and witness and spousal-consent signature status all identify third parties who did not choose the AI client.
• Investor tools need Fund Admin, and the gate is enforced. Carta's connection handshake returns an is_fund_admin_user flag and its own skill stops when it is false.
• Only one RFC 9728 path form works. The root and path-insert forms both 404. A metadata check that tries only the conventional root form would wrongly report no descriptor.
• Carta collects plugin telemetry. Its investors documentation states the plugin collects session context, model configuration and plugin activity to improve reliability, and directs readers to its privacy policy.
• Prompts are unconfirmed. Anthropic's 2026-08-16 snapshot lists an empty prompt_names array, and the gated endpoint blocked our own check.

Frequently asked questions

Sources

• Setting up Carta Plugins — https://docs.carta.com/api-platform/docs/claude-plugins-setup (fetched 2026-08-19 as markdown via the .md suffix the site advertises; page updatedAt 2026-08-18. Source of the three-plugin structure, the shared endpoint, the Owner-only admin gate on Team and Enterprise, the Pro and Max personal path, the Claude Code and ChatGPT Work instructions, the GitHub marketplace route, the troubleshooting notes on Fund Admin data sharing, and the statement that financial data flows through Claude) · retrieved 2026-08-19
• Carta Cap Table guide — https://docs.carta.com/api-platform/docs/carta-cap-table (fetched 2026-08-19; page updatedAt 2026-06-02. Source of the read surface — stakeholders, share classes, voting power, SAFEs, convertible notes, option pools, grants and vesting, 409A valuations, financing history and compensation benchmarks — the read-only statement for this plugin, the modelling capabilities, and the permissions-and-access section) · retrieved 2026-08-19
• Carta Investors guide — https://docs.carta.com/api-platform/docs/carta-investors (fetched 2026-08-19; page updatedAt per site header. Source of the read-only statement, the SELECT-only warehouse query description, the Form ADV point-in-time and entity-coverage limitations, the local-Excel-output statement, the firm-scoping statement, and the tearsheet and query-performance caveats) · retrieved 2026-08-19
• Carta CRM guide — https://docs.carta.com/api-platform/docs/carta-crm (fetched 2026-08-19; page updatedAt 2026-07-28. Source of the read-and-write statement, the create-and-update record list, the firm-scoping statement, and the confirmation-before-ambiguous-writes statement) · retrieved 2026-08-19
• Carta public plugin sources — https://github.com/carta/plugins (fetched 2026-08-19 via the GitHub API and a shallow clone; repository last updated 2026-08-20. Source of the dispatcher evidence, the carta-issuance and carta-witness-signatures skills, the context-versus-corporation rule and its incident log, the inference-first firm selection guidance, the show-and-confirm instruction, the is_fund_admin_user gate, the AI computation gate, the 24-hour handshake cache, and the telemetry statement) · retrieved 2026-08-19
• Live RFC 9728 protected-resource descriptor — https://mcp.app.carta.com/.well-known/oauth-protected-resource/mcp (fetched 2026-08-19; 200, application/json, 295 bytes, eight scopes of which three are readwrite_. The root form and the path-insert form both return 404 with a 9-byte text/plain body) · retrieved 2026-08-19
• Live RFC 8414 authorization-server metadata — https://mcp.app.carta.com/.well-known/oauth-authorization-server (fetched 2026-08-19; 200, issuer https://mcp.app.carta.com/, registration_endpoint, revocation_endpoint, PKCE S256, two grant types, client_id_metadata_document_supported, and a scopes_supported list identical to the protected resource's. The path-append form and /.well-known/openid-configuration both 404) · retrieved 2026-08-19
• Anonymous initialize probe of https://mcp.app.carta.com/mcp (2026-08-19; HTTP 401 with a JSON invalid_token body and a www-authenticate challenge naming the path-append resource_metadata URL. No authentication was attempted and no tool was called) · retrieved 2026-08-19
• Carta documentation robots.txt — https://docs.carta.com/robots.txt (fetched 2026-08-19; 200, allows all agents, disallows only edit, login and CDN paths, and carries no Content-Signal directives. A root llms.txt returns 404; the index is published at /api-platform/llms.txt) · retrieved 2026-08-19
• Anthropic connector directory — https://claude.ai/directory/4b463f8b-6db0-4d0e-9693-8b39f37e4447 (snapshot 2026-08-16; source of the nine tool_names, the empty prompt_names array, the "Read" permissions label, the free-text description quoted in summary, the partner tier, the financial-services category and the documentation URL) · retrieved 2026-08-16