Tool Reference

The Bland MCP server exposes the tools below plus 3 prompts. All of them are scoped to one organization: the one that owns the API key, or the one you chose when you signed in with Bland. Access describes each tool’s annotation: Read tools never change anything and clients can call them freely; Write and Destructive tools have side effects (Destructive ones spend money, touch live traffic, or delete things) and are confirmation-gated: the tool asks for an explicit confirmation argument before executing. Over a sign-in connection, Read tools need the mcp:read scope and Write and Destructive tools need mcp:write.

Calls

Tool	Access	Description
`create_call`	Destructive	Place an outbound call to a phone number, optionally with a task or pathway. Omit `voice` to use Karen.
`get_call_log`	Read	Fetch a single call: its transcript, recording, summary, and metadata.
`wait_for_call`	Read	Wait up to 45 seconds for a call to finish, then return its call log. Use it instead of polling `get_call_log`.
`stop_call`	Destructive	End a call in progress, or cancel one that is still queued.
`list_voices`	Read	List Bland’s curated voices, the recommended set for `create_call`. Any voice the user names still works.

Choose the call resource before invoking a tool

Follow Choose what runs your call: resolve whether the user wants a dashboard agent, persona, or pathway before sending. Honor an established choice, verify the exact ID and environment or version, and ask before changing routes. create_call accepts phoneNumber, task, pathwayId, voice, from, and firstSentence. It does not accept agent_id, agent_version, persona_id, or pathway-version selectors. Do not replace an agent choice with pathwayId just to fit this schema. When available in your connection, use call_bland_api with method: "POST", path: "/v1/calls", and the snake_case request body from Send Call. This supports agent, persona, and version-specific pathway dispatch. For example, after verifying the destination and agent ID:

{
  "method": "POST",
  "path": "/v1/calls",
  "body": {
    "phone_number": "+14155550100",
    "agent_id": "00000000-0000-4000-8000-000000000001",
    "agent_version": "production"
  }
}

Use list_agents, get_agent, and list_agent_environments to verify an agent and its deployment before calling. If the required tool or route is unavailable, explain the limitation and ask the user how to proceed. Do not silently use a persona or pathway instead. After submission, retain the call_id, check the actual source and destination numbers and resource/version association, and verify agent calls under the intended agent’s Conversations view. The create_call result and call-log widget alone do not establish agent attribution. Report a successful call and an attribution problem separately, without placing a duplicate call.

Pathways

Tool	Access	Description
`validate_pathway`	Read	The authoritative Bland pathway compiler. Pass the raw graph you’re about to commit and get structural and runtime-contract errors before saving.
`get_pathway_schema`	Read	The authoritative allowed shape and enums for a structured pathway surface.
`get_pathway_context`	Read	Deep per-node and per-edge execution semantics, the same summaries Bland’s server-side pathway agent uses.

For Bland Agents builds, migrations, snapshot audits, and simulations, use the Norm plugin (norm@bland). The separate bland@bland plugin covers the v1 pathway workspace.

Agents

Full lifecycle for the Bland Agents platform: versioned, branched, and promoted through dev → staging → production environments.

Tool	Access	Description
`create_agent`	Write	Create a new agent plus its dev/staging/production environments.
`list_agents`	Read	List the organization’s agents, most recently updated first.
`get_agent`	Read	Fetch one agent by id, including which version each environment is pinned to.
`rename_agent`	Write	Rename an agent.
`delete_agent`	Destructive	Soft-delete an agent.
`save_agent_version`	Write	Save an agent’s config as an immutable version snapshot.
`list_agent_versions`	Read	List an agent’s saved version history (metadata only), newest first.
`get_latest_agent_version`	Read	Fetch the latest saved version including its full config snapshot.
`get_agent_version`	Read	Fetch a specific published version by semver, including its full snapshot.
`list_agent_environments`	Read	List dev/staging/production environments and the semver each is pinned to.
`publish_agent`	Destructive	Save a snapshot as a new version, mint the next semver, and pin staging to it.
`promote_agent`	Destructive	Pin production to whatever staging currently points at.
`list_agent_variables`	Read	List an environment’s variables for an agent.
`set_agent_variable`	Destructive	Create or update one environment variable (supports secrets).
`delete_agent_variable`	Destructive	Delete one environment variable.
`list_agent_branches`	Read	List an agent’s open branches (isolated dev workspaces).
`create_agent_branch`	Write	Branch off the agent’s current dev head.
`delete_agent_branch`	Destructive	Abandon an open branch (soft delete; its versions stay).
`rebase_preview_agent_branch`	Read	Dry-run a branch merge: three-way diff reporting what auto-merges and what conflicts.
`rebase_agent_branch`	Destructive	Rebase a branch onto the dev head with per-conflict resolutions.
`merge_agent_branch`	Destructive	Merge an open branch into dev (fast-forward only) and archive the branch.

Analytics

Tool	Access	Description
`query_analytics`	Read	Structured call-analytics queries (metrics, filters, dimensions, and time ranges) returning aggregated results.

Evals

Tool	Access	Description
`list_eval_agents`	Read	List the organization’s eval agents (LLM judges) with their latest-run summaries.
`get_eval_agent`	Read	Fetch one eval agent: criterion prompt, scoring levels, targets, weight, modality.
`get_eval_run`	Read	Fetch one eval run and its per-atom verdicts.
`create_eval_run`	Destructive	Enqueue scoring of a batch of calls against eval agents or a saved workbench setup.

Documentation

Tool	Access	Description
`search_bland_docs`	Read	Search docs.bland.ai; returns matching pages with title, URL, and description.
`get_bland_doc`	Read	Fetch the full markdown of one docs page (e.g. `api-v1/post/calls`).
`search_bland`	Read	Search the Bland knowledge base for guides, code examples, and API references.
`query_docs_filesystem_bland`	Read	Read-only shell-like queries over a virtual filesystem of all docs pages and OpenAPI specs.
`submit_feedback`	Read	Report a documentation problem to the docs team.

API passthrough

For anything without a dedicated tool, the passthrough pair gives agents the entire Bland REST API:

Tool	Access	Description
`bland_api_get`	Read	GET any Bland API endpoint with the caller’s key (e.g. `/v1/calls` with query params).
`call_bland_api`	Destructive	Call any mutating Bland endpoint (POST/PUT/PATCH/DELETE). Confirmation-gated.

Setup

Tool	Access	Description
`get_bland_mcp_setup`	Read	Returns connection guidance for the Bland MCP server.

Prompts

Clients that support MCP prompts (Claude Code surfaces them as slash commands) also get:

Prompt	What it does
`bland_pathways`	Loads Bland’s pathway node/edge doctrine into the session.
`build_pathway`	Guided create → validate → test → publish workflow for a new pathway (takes a `goal`).
`edit_pathway`	Guided safe-edit workflow for an existing pathway (takes `pathway_id` and `change`).

Docs for agents: llms.txt

Core Bland Features

AI Agents & MCP

Basic Tutorials

Advanced Features

Platform Information

SDKs & Tools

Bland Enterprise

Tool Reference

Calls

Choose the call resource before invoking a tool

Pathways

Agents

Analytics

Evals

Documentation

API passthrough

Setup

Prompts

​Calls

​Choose the call resource before invoking a tool

​Pathways

​Agents

​Analytics

​Evals

​Documentation

​API passthrough

​Setup

​Prompts

Calls

Choose the call resource before invoking a tool

Pathways

Agents

Analytics

Evals

Documentation

API passthrough

Setup

Prompts