Official Swarme capability MCP at https://swarme.io/mcp. Streamable HTTP. serverInfo: name swarme, title Swarme Capability MCP Alpha, version 2026.07.15-wallet-alpha. 22 control-plane tools (search/describe/quote/run), not 1940 MCP functions. Catalog ~1940 tools (560 free / 1380 premium). Quote before run. Many describe_only. Homepage: https://swarme.io Docs: https://swarme.io/developers
Scored across 25 tools
Most tools are clearly grouped by domain (account, ai, vault, discover), but the agent_* and tool_* namespaces overlap: agent_run/tool_run both execute capabilities, agent_plan overlaps with tool_run's planning mode and tool_quote, and agent_status/tool_status both report on capability runs. Descriptions clarify the Hermes-vs-direct distinction, so an agent can usually pick correctly, but the boundaries are not crisp.
All tools share the swarme_ prefix and use snake_case with a domain segment followed by an action or resource noun. There is no strict verb_noun pattern, and a few are noun-only, but the naming is predictable and readable. Minor inconsistencies like capabilities_search vs capability_describe singular/plural keep it from a perfect score.
25 tools is at the upper boundary of what feels reasonable and touches many subdomains including account, agent, AI, capabilities, discover, experiments, uploads, and vault. Each tool has a role, but the size is heavy and there is noticeable duplication of execution/status patterns across agent and tool. It is not egregious, but it is borderline.
The main workflows are covered end-to-end: capability search/describe/quote/run/status/cancel/artifacts, AI model list/run/status, experiment create/status/select winner, and vault summary/fill/document/field operations. The main gaps are lack of list operations for runs/jobs/experiments and no cancellation/delete for AI jobs or vault fields, but these can be worked around. No critical dead end is apparent.
Return account credit, metered usage, and API client spend-limit status for the authenticated Swarme machine client.
Inputs: no named parameters. Returns: object
Use Hermes to choose a Swarme capability, prepare normalized input, and return quote/execution metadata without running the tool.
Inputs: task (required), files. Returns: object
Use Hermes to choose a Swarme capability and run it when the capability, input, billing, and policy gates allow machine execution.
Inputs: task (required), files. Returns: object
Return capability run status for a Hermes-planned run.
Inputs: run_id (required). Returns: object
List curated AI models available for metered human and machine workflows.
Inputs: sort, limit. Returns: object
Enqueue a curated AI model using the authenticated account credit balance.
Inputs: slug (required), prompt (required), options. Returns: object
Return status and sanitized output for one AI model job.
Inputs: job_id (required). Returns: object
Search the Swarme tool capability registry by keyword and category.
Inputs: limit, query, category. Returns: object
Return the schema for one Swarme capability.
Inputs: slug (required). Returns: object
Read one public or unlisted owner-verified Passport by UUID.
Inputs: passport_id (required). Returns: object
Search public, active, owner-verified Swarme Passports. Unlisted and private records are excluded.
Inputs: host, page, type, limit, +1 more. Returns: object
Run 2–6 compatible models against one canonical brief with a credit budget.
Inputs: task (required), input (required), models (required), blind, +2 more. Returns: object
Select one completed experiment result as the winner and record a bounded score.
Inputs: item_id (required), experiment_id (required), score. Returns: object
Return sanitized model outputs, latency, credit use, scores, and experiment state.
Inputs: experiment_id (required). Returns: object
Return artifact manifest metadata for a Swarme capability run.
Inputs: run_id (required). Returns: object
Cancel a queued run or request cooperative cancellation for a running worker-backed Swarme capability run.
Inputs: run_id (required), reason. Returns: object
Prepare a price/access quote before running a Swarme capability.
Inputs: slug (required), input. Returns: object
Run or plan a Swarme capability. Browser-first tools return a client execution plan.
Inputs: slug (required), input, quote_id. Returns: object
Return the current status, cost, quote, and output metadata for a Swarme capability run.
Inputs: run_id (required). Returns: object
Create a signed upload reservation for a machine-native file capability.
Inputs: slug (required), filename (required), byte_size (required), mime_type (required), +1 more. Returns: object
Return sanitized Vault document inventory metadata without document contents, artifact bytes, share tokens, or direct download links.
Inputs: limit, include_templates. Returns: object
Create or update one Vault field using an owner-issued temporary grant. Responses contain masked metadata only.
Inputs: field_label (required), field_value (required), vault_grant (required), field_id, +4 more. Returns: object
Fill a bounded template using Vault data. Requires an owner-issued temporary grant and returns only sanitized document, billing, and one-use artifact-delivery metadata.
Inputs: vault_grant (required), template_body (required), quote_id, output_format, +2 more. Returns: object
Return a bounded Vault document-fill quote with wallet, spend, provider-cost, and quote-lock policy metadata.
Inputs: template_body (required), output_format. Returns: object
Return profile metadata, grouped field counts, and masked field previews for the authenticated Vault. Decrypted values are never returned.
Inputs: limit, include_fields. Returns: object
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Add one secure layer between your agents and this server.