generated: '2026-08-12'
method: searched
status: published
source: >-
  https://developer.flora.ai/mcp and https://developer.flora.ai/mcp/tools
  (FLORA's own developer site), corroborated by a live probe of
  https://agents.flora.ai/mcp and its RFC 8414 / RFC 9728 discovery documents.
probe:
  url: https://agents.flora.ai/mcp
  method: POST
  body: '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
  http_status: 401
  www_authenticate: >-
    Bearer realm="OAuth", error="invalid_token",
    error_description="Missing or invalid access token",
    resource_metadata="https://agents.flora.ai/.well-known/oauth-protected-resource/mcp"
  fetched: '2026-08-12'
  note: >-
    tools/list is OAuth-gated, so the live inputSchema for each tool could not be
    read anonymously. The tool names, arguments and constraints below are taken
    verbatim from FLORA's published tools reference, not guessed. Full schemas
    require an authenticated introspection.
server:
  name: flora
  vendor_operated: true
  transport: http           # streamable HTTP; "http" in client configs
  url: https://agents.flora.ai/mcp
  sse_url: null
  protocol: Model Context Protocol
  deprecated_endpoints:
    - url: https://app.flora.ai/mcp
      note: >-
        FLORA's product docs state "the legacy /mcp beta endpoint has been
        deprecated" and direct clients to the /api/v1-backed install URL on the
        developer site. Probed 2026-08-12: 404.
      status: 404
auth:
  methods: [oauth2]
  oauth2:
    version: OAuth 2.1
    authorization_server: https://agents.flora.ai
    authorization_endpoint: https://agents.flora.ai/authorize
    token_endpoint: https://agents.flora.ai/token
    registration_endpoint: https://agents.flora.ai/register
    revocation_endpoint: https://agents.flora.ai/token
    grant_types: [authorization_code, refresh_token]
    pkce: [S256, plain]
    dynamic_client_registration: true
    scopes_supported: null
    scopes_note: >-
      The authorization-server metadata does not enumerate scopes_supported. The
      docs describe the grant in terms of the signed-in user's existing FLORA
      role: the MCP connection inherits exactly what that user can do in the
      workspace, and cannot manage billing or workspace members at all.
    discovery:
      authorization_server: well-known/flora-fauna-agents-oauth-authorization-server.json
      protected_resource: well-known/flora-fauna-agents-oauth-protected-resource.json
      protected_resource_mcp: well-known/flora-fauna-agents-oauth-protected-resource-mcp.json
  api_key_alternative:
    supported: false
    note: >-
      Unlike most MCP servers in the catalog, FLORA does NOT accept its REST
      sk_live_ API key on the MCP endpoint. The docs are explicit that the two
      auth paths are independent: API keys for server-side automation, OAuth for
      interactive agents.
tool_count: 2
tool_model: code-execution
tool_model_note: >-
  FLORA deliberately does not wrap one endpoint per tool. It exposes two tools and
  lets the agent write TypeScript against a pre-authenticated @flora-ai/flora SDK
  client inside `execute`. The effective capability surface is therefore the SDK
  method surface (see mcp/flora-fauna-tool-crosswalk.yml), not the tool count.
tools:
  - name: search_docs
    category: discovery
    description: >-
      Searches the @flora-ai/flora SDK documentation to find the right method, its
      parameter shapes, and usage examples. The agent calls this before writing
      code when unsure of a method name, argument structure, or return type.
    arguments:
      - {name: query, type: string, required: true, description: Natural-language question or method name to look up}
      - {name: language, type: string, required: false, description: 'Target language, e.g. "typescript"'}
  - name: execute
    category: execution
    description: >-
      Runs a TypeScript snippet in a sandbox against a pre-authenticated
      @flora-ai/flora client. The agent defines `async function run(client) { ... }`
      and the server invokes it, returning the value plus any console.log output.
    arguments:
      - {name: code, type: string, required: true, description: 'TypeScript source defining async function run(client). Exact argument name is not published; schema requires authenticated tools/list.'}
    constraints:
      state_persistence: none — each execute call is isolated
      http_request_timeout: 30s
      total_execution_timeout: ~5 minutes
client_namespacing:
  claude_code: [flora:execute, flora:search_docs]
  cursor: [flora.execute, flora.search_docs]
install:
  claude_code: claude mcp add --transport http flora https://agents.flora.ai/mcp
  generic_json: '{"mcpServers":{"flora":{"type":"http","url":"https://agents.flora.ai/mcp"}}}'
  stdio_bridge: npx -y mcp-remote https://agents.flora.ai/mcp
  documented_clients:
    - Claude (web + Desktop connector)
    - Claude Code
    - OpenAI Codex
    - Cursor
    - VS Code
    - Hermes Agent (Nous Research)
    - OpenClaw
    - any client supporting streamable HTTP + OAuth 2.1
entitlement:
  required_plan: Starter
  note: >-
    FLORA's pricing docs state the Free plan has "no FLORA API or MCP"; both start
    on Starter ($18/mo). MCP runs bill in USD against the same workspace budget as
    the app, at the same published model rates.
errors:
  - {code: invalid_token, http: 401, meaning: OAuth token expired or revoked; reconnect the client}
  - {code: insufficient_credits, http: 402, meaning: workspace USD balance too low to start the run}
  - {code: forbidden, http: 403, meaning: the signed-in FLORA user lacks permission for the operation}
  - {code: rate_limited, http: 429, meaning: workspace rate limit; the docs say it is rare interactively and common in batches}
registry_listings: []
registry_note: >-
  No listing was found in a public MCP registry, the Claude connector directory,
  or the ChatGPT apps directory as of 2026-08-12. Distribution is via the
  provider's own install pages.