Problem

integrations/beads-mcp/src/beads_mcp/models.py:23 defines:

DependencyType = Literal["blocks", "related", "parent-child", "discovered-from"]

But the bd CLI emits ~19 dependency types (see internal/types/types.go:769-805):

blocks, parent-child, conditional-blocks, waits-for, related, discovered-from, replies-to, relates-to, duplicates, supersedes, authored-by, assigned-to, approved-by, attests, tracks, until, caused-by, validates, delegated-from

Any mcp__beads__show() call on an issue with a non-whitelisted dep edge crashes with pydantic literal_error. Repro from a real session today:

beads - show (MCP)(issue_id: "tn-b77d", brief: true, brief_deps: true)
  ⎿  Error: 1 validation error for Issue
     dependencies.0.dependency_type
       Input should be 'blocks', 'related', 'parent-child' or 'discovered-from'
     [type=literal_error, input_value='relates-to', input_type=str]

brief=true doesn't help — BriefDep reuses the same Literal. The CLI (bd show) is unaffected because it doesn't go through Pydantic.

This blocks any agent workflow that uses mcp__beads__show() over a knowledge graph that has relates-to, replies-to, duplicates, supersedes, etc. edges. PR #1469 added CLI display support for relates-to back in February — the MCP layer just never caught up.

Related: #1750 (MCP Pydantic validation error on list when issues have dependencies — same class of bug, different call site).

Fix

Change DependencyType from a Literal to a plain str, mirroring how IssueStatus and IssueType are already handled in the same file (lines 21-22). The bd CLI is the source of truth and validates dep types itself (DependencyType.IsValid() in internal/types/types.go:805); the MCP layer doesn't need a redundant whitelist that drifts from the CLI.

Also updated the user-facing docstrings in server.py and tools.py to stop claiming only 4 dep types exist — they now mention the common four and point to internal/types/types.go for the full list.

Verification

Synthesizing an Issue with a relates-to LinkedIssue now validates cleanly:

from beads_mcp.models import LinkedIssue, Issue
import datetime
now = datetime.datetime.now(datetime.timezone.utc)
linked = LinkedIssue(
    id='tn-test', title='test', status='open', priority=2, issue_type='task',
    created_at=now, updated_at=now, dependency_type='relates-to'
)
Issue(
    id='tn-parent', title='parent', status='open', priority=2, issue_type='task',
    created_at=now, updated_at=now, dependencies=[linked]
)
# OK — no validation error

Test suite: 141 unit tests pass, 7 model/dep-specific tests pass. Note: there are 4 pre-existing failures in tests/test_compaction_config.py that hardcode /Users/stevey/src/dave/beads/... paths — unrelated to this change, fail on any non-Steve machine.

Why a str and not an expanded Literal?

I considered listing all 19 types in the Literal, but:

1. New dep types get added to internal/types/types.go over time (e.g., delegated-from is recent). A Literal will drift again.
2. The MCP layer can't validate semantic correctness anyway — only the Go-side IsValid() method knows the live set.
3. IssueStatus and IssueType already made the same call (and for the same reason — custom statuses/types via bd config set), so this is internally consistent.

If you'd prefer the literal approach for type safety / IDE autocomplete, happy to switch.