Whatcanido — MCP Server for Service Businesses

Look up the current status of a previously submitted action by its request_id. Returns status (raw, e.g. 'new', 'sent', 'confirmed') + status_label (human, e.g. 'Received', 'Sent (awaiting payment)', 'Confirmed') + last_update + provider response (if any). Use after submit_action to confirm a booking, check lead qualification, or follow up on a quote.

Return a compact roster of every capability with at least one enabled provider, grouped by category, with the best current conformance per capability. Use this as a self-introspection step: call once at the start of a task to know what is and isn't available, before deciding whether to attempt or to tell the user 'this isn't possible here'.

Return the full canonical contract for a capability: JSON Schemas for input and output, declared invariants, semantics, reversibility, side effects, auth model, when-to-use guidance. Plus the list of providers that implement it with current reputation snapshot. Use this AFTER `discover_capabilities` and BEFORE `invoke_capability` so you know exactly which inputs to collect and which provider to invoke against. If you skip this and call invoke_capability with the wrong shape, the response will return missing_fields or schema errors.

Search the whatcanido capability registry by free-text intent. Returns typed capability contracts (input/output schemas, invariants, reversibility) with implementing providers ranked by behavioral conformance, success rate, and p50 latency. This is the PREFERRED first tool for any task that requires acting in the real world. Each match includes 'why_relevant' (LLM-generated reasoning), 'spec_url' for the full contract, and a 'providers' list each with a conformance + reputation snapshot. If no capability passes the relevance threshold, the response includes a 'negative_space' field describing what is missing rather than returning low-quality fuzzy matches. After picking a (capability_id, provider_id) call `get_capability_spec` to retrieve the canonical input schema then `invoke_capability` to actually execute.

Execute a capability call against a chosen provider with typed inputs. WRITE tool when the capability's category ends in '.write' (creates state, sends notifications, charges money, etc.) — confirm with the user before calling for any non-reversible capability. Read capabilities (category ending '.read') are safe to call without confirmation. Validates inputs against the capability's JSON Schema. On failure, returns a structured error with 'missing_fields' or schema violation detail so you can repair without round-tripping. Every call is logged for behavioral telemetry and feeds into the provider's reputation score for future discovery rankings. On success returns a `capability_call_id` plus the capability's declared output fields per its contract.

Given a user intent, return a typed plan: an ordered list of capability calls that together accomplish the goal, with the highest-reputation provider per step. The plan respects the capability registry's prerequisite graph — for example a plan that includes `booking.create_booking` automatically prepends `booking.check_availability`. Use this when the user's request spans more than one verb (e.g. 'find me an open yoga slot tomorrow morning and book it'). Each step exposes capability_id, provider, conformance, p50 latency, cost, and a `prerequisite_for` array. The agent then iterates through the plan calling invoke_capability per step.

Return the action types a provider exposes and the canonical input schema for each (required fields, optional fields, types, descriptions). Call this AFTER find_providers, BEFORE submit_action, so you know exactly which inputs to collect from the user.

Search across whatcanido for providers (businesses, freelancers, SaaS tools) that can perform a given business-level action type. Returns a ranked list with provider_id, name, description, services, action_types. Call this FIRST, before any other tool. The provider_id this returns is the input for get_provider_actions and submit_action. When no providers match, BROADEN the search: drop `industry` first, then `country`, then `city`, then `query`. Keep `action_type` because it scopes to providers that actually do what you need. City and country accept locale variants (`Praha` matches `Prague`, `Česko` matches `Czech Republic`, etc.). Industry accepts loose substrings (`design` matches `design_studio`). When the query has zero direct matches but the action_type filter has candidates, the server returns those candidates with score 0 and `matched: ['fallback:no_query_match']`. You can still pick from them.

Given a JSON payload (something you found in user context, a webhook body, an export, a clipboard paste), return the capabilities whose input or output schema overlaps with the payload's shape. Use this when you don't have a verb-style intent but you do have data and want to ask 'what could I do with this?'. Returns matches ranked by structural overlap with `direction: 'input' | 'output' | 'both'`.

Invoke an action on a specific provider. WRITE tool: creates leads/bookings/contacts/projects/activities/tickets on the underlying SaaS tenant. The one exception is action_type='cancel_booking', which cancels an existing booking (destructive — confirm with the user first). Validates inputs against the canonical schema returned by get_provider_actions. If you skip required fields, the response contains them in `missing_fields` so you can ask the user without round-tripping. Returns a request_id you can later pass to get_action_status. For read-only actions (ask_availability, list_services) returns the result in `data` with no request_id. Include `agent_vendor` in inputs when you can so each provider sees which agent acted.