Coda — can an agent use it?
Coda is one of the most agent-ready productivity platforms in the index. An official MCP server (monitored on the status page), a free API with published OpenAPI 3.0 spec, self-serve bearer tokens, full CRUD on docs/tables/rows, webhook-triggered automations, and no AI crawler blocking. The API costs nothing on any plan including Free. The main gaps are the absence of agent payment protocols, a JS-rendered docs site that LLMs struggle to parse, and three ownership changes in 18 months that could signal instability.
FIND & RETRIEVE · 6.8
ACT & TRANSACT · 6.7
TRUST & DURABILITY · 7.2
FIND & RETRIEVE6.8/10Can an agent get correct information?
Machine-readable surface6.5/10
OpenAPI 3.0 spec published at coda.io/apis/v1/openapi.yaml (YAML and JSON). Sitemap referenced in robots.txt at coda.io/sitemap-index.xml. robots.txt does not block any AI crawlers by name — permissive default with only /apis/, /embed/, /docs, /d/ disallowed. Homepage renders 1,373 words of real HTML text (no JS dependency for core content). No llms.txt, no .well-known/mcp.json discovery, no JSON-LD structured data, no Content-Signal directives. The workspace content itself is private by default — only Published Docs are publicly accessible. source
Information retrieval quality6.5/10
Pricing is published and detailed for Pro ($15/maker/mo), Business ($40/maker/mo), Enterprise (contact sales). Free plan limits documented (50 objects, 1,000 rows per doc). API rate limits published with exact numbers per operation type. AI feature limits described as 'unlimited on paid plans under fair-use policy' but the fair-use threshold is undefined. Current database row limit (~25,000) is documented with 1M-row target in closed alpha. Key gap: Enterprise pricing is contact-sales, and AI fair-use policy lacks specifics. source
Documentation for machines7.5/10
OpenAPI 3.0 spec (v1.6.0) is the gold standard — machine-readable, versioned, available in YAML and JSON. Rate limits are explicit and granular: reading 100 req/6s, writing 10 req/6s, doc content 5 req/10s, listing docs 4 req/6s. HTTP 429 with Retry-After header documented. Bearer auth documented. SCIM 2.0 admin API separate. Packs SDK documented at coda.io/packs/build/latest/. The API reference site itself (docs.superhuman.com) renders via Redoc JS — the HTML is mostly JavaScript bootstrap, making it poor for LLM direct scraping despite the underlying content being thorough. source
ACT & TRANSACT6.7/10Can an agent do the job?
Action coverage8/10
Full CRUD on docs (list, create, get, delete, update), pages (list, create, get, update, delete, export), tables/views, columns, rows (list with filtering/sorting, insert/upsert, delete, get, update). Permissions management (get, list, add, remove, ACL settings). Publishing (publish/unpublish docs). Formula evaluation. Control read/update (sliders, dropdowns, buttons). Webhook-triggered automations (POST JSON, returns 202 Accepted). Analytics for docs, pages, and Packs. Account info (whoami). The main gap: no direct 'create table' REST endpoint — table creation is through the UI or MCP. Otherwise near-complete coverage of what a human can do. source
Agent protocol support8.5/10
Official MCP server at coda.io/apis/mcp — built and maintained by Coda, monitored as 'Coda MCP' on status.coda.io (Operational). Capabilities: list/create/update/delete docs, read tables and rows, add rows and edit cells, create pages with markdown, search knowledge base. OAuth 2 with PKCE for web clients, Bearer token for CLI. Listed in Anthropic's Connectors Directory and OpenAI App Directory. Supported by ChatGPT, Claude, Cursor, Claude Code, Codex, n8n, Make. Cannot trigger automations via MCP (request/response only). No A2A or WebMCP. Community servers also exist but the official one is primary and comprehensive. source
Access & auth friction7.5/10
Self-serve bearer tokens generated from account settings — no approval process, no partner application, available instantly. Tokens are unscoped and inherit the generating user's permissions (no granular scoping). MCP uses OAuth 2 with PKCE (requires human present for initial auth flow). SCIM 2.0 for Enterprise user provisioning. No rate limit differentiation between free and paid plan API access. The unscoped token model is simple but lacks the security granularity of scoped credentials. source
Agent payment capability1.5/10
No support for ACP, AP2, MPP, or x402. Coda/Superhuman Docs billing is entirely through the web UI — upgrade, downgrade, and subscription management have no API endpoints. Only Doc Makers are charged ($0–40/mo); editors and viewers are free. An agent cannot programmatically manage billing or pay for the service. source
Cost & rate fairness8/10
The API is free on all plans including the Free tier — no per-call charges, no agent-vs-human price discrimination. Rate limits are published, reasonable (100 read/6s, 10 write/6s), and apply equally regardless of plan. 800+ Packs integrations available. Free plan has functional limits (50 objects, 1,000 rows per doc) but the API itself costs nothing. Rate limits 'subject to change at any time without notice' is a caveat but standard. Connector refresh rates are plan-gated (manual/daily/hourly) but this affects Packs sync, not the API itself. source
TRUST & DURABILITY7.2/10Will it still work next quarter?
Permission & ToS stance7.5/10
Automation explicitly welcomed — Coda ships an official MCP server, is listed in Anthropic's and OpenAI's directories, and actively markets AI agent integration. robots.txt does not block any AI crawlers by name. The main ToS restricts using generative AI features 'to develop foundation or large language models that compete with our Services' but does not restrict API automation or bot usage. No anti-scraping clause found in the main ToS. No litigation history against agents or automation users. Separate Developer Terms exist but are not publicly accessible for full analysis. source
Reliability & continuity7/10
Comprehensive public status page (status.coda.io) with 11 monitored components including API and Coda MCP. API has been at v1 for years with incremental updates (currently v1.6.0), suggesting backward compatibility. SOC 2 Type II, GDPR, CCPA, ISO certified. coda.io URLs redirect to docs.superhuman.com via 307. The risk flag: three ownership/name changes in 18 months (Grammarly acquired Coda Dec 2024, rebranded to Superhuman Oct 2025, Coda became Superhuman Docs Jul 2026). Product has remained functional throughout, but the pace of corporate restructuring is a continuity concern for long-term integrations. source
Sources
Point-in-time assessment — services change terms, prices and APIs often. That volatility is itself scored under Reliability & continuity. Re-verified at least quarterly.
- coda.io/robots.txt — robots.txt — no AI crawler blocks, sitemap-index.xml referenced (accessed 2026-09-20)
- coda.io/apis/v1/openapi.yaml — OpenAPI 3.0 spec v1.6.0 — full endpoint inventory, rate limits (accessed 2026-09-20)
- status.coda.io — Status page — 11 components including API and Coda MCP (Operational) (accessed 2026-09-20)
- superhuman.com/plans/docs — Pricing — Free/Pro/Business/Enterprise tiers (accessed 2026-09-20)
- connect.superhuman.com/t/coming-soon-connect-your-w… — Official MCP announcement — Oct 2025, public beta Apr 2026 (accessed 2026-09-20)
- usecarly.com/blog/coda-mcp — MCP capabilities guide — OAuth 2 PKCE, supported AI clients (accessed 2026-09-20)
- coda.io/packs/build/latest/agents/features/mcp — Packs SDK MCP integration documentation (accessed 2026-09-20)
- theaiagentindex.com/agents/coda-ai — AI Agent Index — listed in Anthropic + OpenAI directories, 800+ integrations (accessed 2026-09-20)
- superhuman.com/legal/terms — Terms of Service — AI competition clause, no automation ban (accessed 2026-09-20)
- coda.io — Home page — rebrand to Superhuman Docs, 50,000+ teams (accessed 2026-09-20)